Content uploaded by Matthias Volk
Author content
All content in this area was uploaded by Matthias Volk on Dec 17, 2020
Content may be subject to copyright.
Received September 14, 2020, accepted September 24, 2020, date of publication October 1, 2020, date of current version October 22, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.3028127
Identifying Similarities of Big Data Projects—
A Use Case Driven Approach
MATTHIAS VOLK , (Graduate Student Member, IEEE), DANIEL STAEGEMANN,
IVAYLA TRIFONOVA, SASCHA BOSSE, AND KLAUS TUROWSKI
Faculty of Computer Science, Otto von Guericke University Magdeburg, 39106 Magdeburg, Germany
Corresponding author: Matthias Volk (matthias.volk@ovgu.de)
ABSTRACT Big data is considered as one of the most promising technological advancements in the last
decades. Today it is used for a multitude of data intensive projects in various domains and also serves as
the technical foundation for other recent trends in the computer science domain. However, the complexity
of its implementation and utilization renders its adoption a sophisticated endeavor. For this reason, it is not
surprising that potential users are often overwhelmed and tend to rely on existing guidelines and best practices
to successfully realize and monitor their projects. A valuable source of knowledge are use case descriptions,
of which a multitude exists, each of them with a varying information density. In this design science research
endeavor, 43 use cases are identified by conducting a thorough literature review in combination with the
application and adaption of a corresponding template for big data projects. By a subsequent categorization,
which is performed by identifying and employing a hierarchical clustering algorithm, nine different standard
use cases emerge, as the contribution’s artifact. This provides decision-makers with an initial entry point,
which can be utilized to shape their project ideas, not only by identifying the general meaningfulness of their
potential big data project but also in terms of concrete implementation details.
INDEX TERMS Big data, use case analysis, clustering, categorization, literature review, design science
research.
I. INTRODUCTION
Due to the ever-growing amount of data produced and cap-
tured by humanity [1], [2], the ability to analyze and subse-
quently use the contained information has gained a widely
acknowledged significance in today’s society [3]. While the
usage of data, in general, is no new concept, the prevailing
‘‘data deluge’’ poses new challenges that overstrain tradi-
tional technologies and demand new solutions [4], leading
to the term ‘‘big data’’. Even though there is no unified
definition of the term, the approach by the National Insti-
tute of Standards and Technology (NIST) belongs to the
most common ones. It states that ‘‘Big Data consists of
extensive datasets – primarily in the characteristics of vol-
ume, velocity, variety, and/or variability – that require a
scalable architecture for efficient storage, manipulation, and
analysis’’ [5]. These data characteristics describe the nature
of the data to be processed and, thus, are often recognized
as crucial influence factors when it comes to the planning
The associate editor coordinating the review of this manuscript and
approving it for publication was Fanbiao Li .
and realization of big data projects. The volume, for instance,
describes the amount of data to be stored, managed, and
processed [5]. In turn, the variety of the data either represents
the heterogeneity of the structure [6], [7], but sometimes
also the origin [5]. The same differentiated view exists for
the velocity that either addresses the speed with which the
data is incoming or the time for its processing [5], [7].
The variability refers to changes of the dataset, for instance,
in terms of the structure, rate of the data flow or the size
of the data [5]. Together, those characteristics are covered
under the umbrella term of the fours Vs of big data, with each
depicting the abbreviation of one characteristic. Besides those
known core characteristics, many others emerged in recent
years, such as the value of the data or the veracity [5], [8].
To handle the additional challenges, posed by those Vs,
compared to traditional workloads, numerous new strategies,
tools, and systems have been introduced in recent years.
Nowadays, those technologies and concepts are applied to
a wide array of domains, comprising, but not being limited
to, mobility [9], smart cities [10], media distribution [11],
healthcare [12], sports [13], education [14] and business [15].
VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ 186599
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
Furthermore, also the positive effects of the utilization of big
data technologies have been shown [16], [17], emphasizing
the importance of such efforts.
However, implementing and operating those systems is a
sophisticated endeavor with many possible pitfalls [18], [19],
which could diminish the benefits or even result in nega-
tive outcomes [20]. By considering the shortage of qualified
experts in this domain and the concurrent demand [21], [22],
independent from the actual size of the enterprise, it appears
to be reasonable to support concerned decision-makers and
technicians. This applies especially to fundamental tasks
like the design of the underlying technical architecture [23].
Hence, a thorough description of corresponding use cases
could facilitate the realization of those kinds of complex
projects by providing a suitable source of information. How-
ever, the biggest deficiency of most of these is the level
of detail. Although every year, numerous contributions are
published, describing the general endeavor, they often omit
information regarding the actual project, the occurred prob-
lems, and the specific implementation. Furthermore, in some
cases, similar methods and paradigms are applied or even
technological considerations made, leading sometimes to a
sole distinction in terms of the main scope. This reduces
transparency for those who want to utilize such cases to
obtain information for similar application scenarios, in terms
of the general meaningfulness, best practices, concrete tech-
nologies, or even specific architecture details. Consequently,
today a multitude of case studies exists, potentially offering
valuable guidance for the realization of big data projects,
but their exploitation still remains a challenge. To uncover
inherent relations between those and thus overcome the out-
lined barriers, the following research question (RQ) shall be
answered in the course of this work:
Which standard big data use cases are revealed, applying
cluster analysis to the corresponding case studies, found in
the literature?
Answering this question and structuring a collection of
successful big data projects, accordingly, could constitute
a valuable resource for the instantiation of future projects
providing a general orientation and guidance but also con-
crete implementation details for specific scenarios. Due to the
presumed similarity of the cases in parts, the formulation of
standard use cases for the structuring of a potential collec-
tion appears to be a suitable solution. As a result, decision-
makers will obtain the opportunity to quickly identify cases,
which are similar to the challenges they are facing and thus,
lean on the existing knowledge base. It also enables them
to classify their projects and identify the needed expertise
in a more systematic and therefore more meaningful way.
Furthermore, those categories could be used as a foundation
for a big data decision support system. This allows for a
comparison of the expected usage scenario with familiar
exemplary cases and the incorporation of gained experiences
in a technology-supported way [24].
Since the first introduction of the most famous big
data technology – Hadoop – in 2005 [25] the maturity
tremendously increased. To take this development into
account, only use cases published between 2015 and
2018 were incorperated, as cases from the period before have
already been sufficiently investigated by other authors, such
as in [26]. Due to the start of this research in 2019 and the
decision to cover only literature of completed years, the end
date 2018 was chosen. Furthermore, to reduce the complexity,
misinterpretations, and effort for further modifications and
extensions, additional document analysis techniques were
applied. While the general approach resembles the work of
Ylijoki and Porras [26], using case study analysis and cluster-
ing, the specific methodology, the analyzed time frame, and
the respective objectives differ. Apart from that, the publi-
cation at hand especially addresses researchers and scientists
concerned with the project feasibility, technology selection as
well as implementation and application details in a big data
context.
A. METHODOLOGY
To find an answer for the initially formulated RQ, multi-
ple methodologies need to be applied in a combined way.
As a general foundation for the realization of this endeavor,
the design science research methodology according to
Hevner et al. [27] was used, providing an artifact as a solu-
tion to the formulated problem. In particular, similarities of
successful big data projects are investigated and standard
use cases are derived as the main artifact of this work.
To further improve the reproducibility and clarity of the
conducted measures, the six-stepped procedure according to
Peffers et al. [28] was followed to ensure, that the develop-
ment of the intended solution is systematically approached.
The first step of this workflow focuses on the brief motivation
and description of the problem. Subsequently, the main objec-
tives are highlighted, which is directly followed by design
and development. As a transition, the theoretical foundation
needs to be investigated and relevant material collected. Since
the main artifact of the research, in the form of standard use
cases, builds upon existent use cases, a structured literature
review [29], [30], as well as a use case analysis were con-
ducted, examining the content of each of the cases in-depth.
Often companies write case studies for advertisement aims,
for example, to win new customers or to present themselves
in social media. On the other hand, they prepare the case
studies as a documentation of their ‘‘best practices’’ [31].
Those describe the decisions made by companies, the reasons
for those decisions, their implementation, and the following
results [32]. Case studies can also be used as a guideline
for subsequent users, having a particular problem, or striving
for a concrete solution. Either way, to provide maxiumum
value, case studies must be written according to pre-defined
standards [33]. Hence, it was presumed that those case stud-
ies describe the usage of big data technologies and the
related processes in their context. To ensure this, additionally,
the comprehensiveness of each of them was checked by using
a modified version of an existing use case template [34].
At this stage, also important features for a later clustering
186600 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
approach, implemented during the subsequent step, are iden-
tified. The following three steps concern the demonstration,
evaluation, and presentation of the artifact.
B. STRUCTURE
Based on the used methodology, the structure is as follows.
Within the first section, an initial overview of the current situ-
ation, as well as the derived research question and main objec-
tives, are presented. Along with this, the used methodology as
well as the structure is introduced. In the subsequent second
section, the conducted structured literature review, resulting
in the collection of the regarded use cases, is described. The
third section thoroughly describes the performed clustering,
each of the clusters and their particularities, as well as the
actual development of the standard use cases. An evaluation
of the obtained results is presented in the fourth section.
In here, the standard use cases are tested on the base of
previously unseen data, at which a categorization of those
is pursued. Within the fifth section, concluding remarks are
given. Apart from a summary, this comprises a discussion and
an outlook on future research.
II. THE STRUCTURED LITERATURE REVIEW
For the identification of relevant big data use cases that have
been published between 2015 and 2018, a structured literature
review is performed. In particular, the methodologies accord-
ing to Levy and Ellis [30] as well as Webster and Watson [29]
were used. To verify the comprehensiveness of the found out
contributions, additionally, an existing use case template was
adopted, which is provided by the NIST [34]. Within the
following section, the review protocol, the used template as
well as the results are meticulously described [35].
A. REVIEW PROTOCOL
To obtain a broad overview of the entire domain, the focus
of the search was not set on a single database. Instead to
‘‘exhaust all sources that contain IS research publications’’
[30, p. 183] the scientific literature database Scopus was
used for the initial keyword search. Although it was noted
by different authors [29], [30] that multiple sources should
be queried, to receive an extensive overview of all of the
relevant articles, due to its comprehensiveness, only the men-
tioned database was used, since most of the widely accepted
literature databases and their relevant articles are listed here,
referring to the source. For these reasons, Scopus serves
more as a kind of a meta-database indexing relevant con-
tributions. Secondly, it was not required to perform alter-
ations on the queried terms and used operators. According
to the targeted domain of interest, the terms ‘‘case study’’,
‘‘use case’’ and ‘‘case description’’ were used in combi-
nation with ‘‘big data’’. Further, to reduce the number of
irrelevant search results, additional inclusion and exclusion
criteria were formulated. Some widely accepted ones are, for
instance, the used language, the publication in a conference,
journal or book, and relevance for answering the formu-
lated research question. Only when all of the aforementioned
inclusion criteria were met, the paper was accepted. However,
sometimes it was noted that some of the contributions did
not encompass as much information as needed. Due to this,
various criteria were formulated, like, if a use case was not
presented very well and did not contain the required key infor-
mation suitably, a paper was rejected. In turn, in a few of the
found out use cases, the information density was very high but
only focusing on the introduction, development, or evaluation
of new technologies. An additional exclusion criterion was
formulated for those cases. The complete collection of all
of the used inclusion and exclusion criteria is summarized
in Table 1. The initial material collection was performed
by applying the described keywords and some of the men-
tioned inclusion criteria directly through the advanced search
mechanics of the literature database. As a result, 2,379 non-
redundant publications were found. Following that, it was
required to check the papers on their actual usability. The
refinement of the material was performed in a two-stepped
procedure. In the first step, the title, abstract, and structure
were checked. This resulted in having 108 relevant contri-
butions. Within the second step, the actual content of the
remaining case studies was investigated. It was noticed that
in most of the cases the content differed strongly, in terms of
the information density.
TABLE 1. Inclusion and exclusion criteria of the structured literature
review.
Only a rough presentation was performed for most of the
use cases, neglecting important descriptive information, such
as about the data or the situation before. To simplify the
mapping of the needed information, the qualitative analy-
sis, and the evaluation of the comprehensiveness for each
use case, the use of a corresponding template was deemed
appropiate. As a basis, the very extensive template provided
by the NIST, covering eight different parts and 57 big data
project-related questions, was modified and used [34]. The
original template was designed by the NIST Big Data Public
Working Group (NBD-PWG) to collect existing use cases.
Due to its overarching purpose, the template as well as the
VOLUME 8, 2020 186601
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
respective categories were continuously validated regarding
their applicability and, thus, compared to the actual content
of the found out contributions.
B. USE CASE TEMPLATE
The basic template comprises a multitude of different aspects.
Apart from the general project description and the situation
before, the relevant big data characteristics, applied tech-
niques, and multiple other information are requested. Due
to the reason that not all of the template’s fields were from
major interest, modifications to the original version were
performed. After an initial scan, during the second step of the
refinement, fields which were not related to the formulated
criteria (cf. Table 1) and not required for the general appli-
cability of big data-related projects were removed (R). This
includes, for instance, the last two parts and questions like
‘‘do you foresee any potential risks from public or private
open data projects?’’ or ‘‘under what conditions do you
give people access to your data?’’ [34]. At the same time,
additional points, like the veracity of the data or privacy-
related information, were newly added (N). All other fields,
meaningful for the found out contributions, were adapted.
An overview of the general content of each template category
is depicted in Table 2, whereas a complete depiction of all
of the made considerations is shown within the appendix
in Table 8. While the first column describes the targeted
content of the field, the latter is focusing on the performed
changes. Either the field of the original use case template was
adapted (A), removed (R) or a new one was added (N).
TABLE 2. Overview of the use case template categories and their general
description (cf. [34]).
In total, 23 of the questions were removed, 27 adapted
without any changes, 7 modified, and 11 newly added. For
instance, within the first part, only minor changes were made,
to prevent misleading interpretations during the investigation
of the use cases. This includes the renaming of the parts,
the deletion of one question, and the addition of the part
advantages of harnessing big data, which was covered by
almost all cases except for one [36]. In the subsequent second
part, additional characteristics were introduced or moved
from the adhering third part, due to the high coverage by
the use case description. In the following, every part was
evaluated regarding the usefulness for the intended compre-
hensiveness check. Consequently, smaller changes were con-
ducted, as they are depicted in the referred Table 8. Notewor-
thy are the last two parts, which were removed completely,
since almost none of the use cases hold information related
to those. While part seven deals with various workflow steps
and respective changes in the data characteristics, the last
one goes into more detail, when it comes to privacy and
security concerns. Although some papers, such as [33], [37],
[38], discussed various issues in detail, the relatively small
number of related cases made those parts not universally
applicable. Eventually, the final template was used to check
the comprehensiveness of each found out use case.
C. RESULTS OF THE LITERATURE REVIEW
After the two-stepped refinement procedure and the compre-
hensiveness check through the used template were finished,
40 different case studies from a keyword-based search in the
academic area remained. To broaden the overview, towards
the practitioner’s perspective, the same criteria, keywords,
and competency questions were used for a search procedure
on industrial case studies. After the keyword search proce-
dure, querying the Google Search Engine, 208 additional
cases were identified.
Through the subsequent observation of the mentioned cri-
teria and the use of the modified template, only three cases
remained. Those are from Lufthansa [39], Dell [33] and
Fujitsu [40]. One of the prevailing factors was the lack of
specific information. Instead, mostly advertisements were
presented to showcase the company’s competence. One pos-
sible reason for this might be the missing motivation to
present critical information, fearing the loss of competitive
advantages. An overview of all of the described steps is
depicted in Figure 1. In total, the material collection resulted
in 43 different cases, which were further used in the course
of this work. Although it was expected that an additional step
in the form of a forward-backward search [29] will increase
the number of promising cases, no new cases were identified.
Most of the cases were published in 2016 (seventeen case
studies), followed by 2017 (twelve case studies), 2018 (nine
case studies) and finally 2015 (five case studies). A general
investigation of the respective application domain reveals that
the chosen case studies are coming from various areas.
Almost 25 percent of all contributions originate from the
healthcare area. Another important area, with 38 percent,
186602 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
FIGURE 1. Conducted steps of the literature review, including the initial
keyword search and the subsequent refinement steps.
is the internet of things (IoT), aiming to realize concepts
like smart city, smart transportation, smart buildings, and
more. Considering the various application areas, the time
scope and the different databases from which the case studies
originate, it can be concluded that those are a representa-
tive sample of the existing projects, regarding the usage of
big data technologies. A complete listing of all of them is
depicted in Table 3. While most of the related databases
are directly addressed, others comprises the databases Taylor
and Francis,Gesellschaft fuer Informatik,ACM,IADIS Por-
tal,Scitepress, and the use cases originating from company
resources. Furthermore, for each paper, a distinct number
is assigned that will be used in the further context of this
work.
TABLE 3. The Results of the Literature Review Mapped to the Respective
Literature Databases.
III. USE CASE ANALYSIS
Prior, a quantitative overview was given and a mapping of
the selected case studies was performed. In the following,
those will be analyzed in a detailed way. Manual approaches
often result in great effort if the analysis has to be repeated or
extended in possible future work. Because of the high number
of found out results, the subjective observation that may come
with this kind of investigation on extensive documents, and to
increase the comprehensibility of this research, a more objec-
tive analytical approach was chosen. In particular, document
clustering was selected, not only to find relevant information
but also as described before, to identify standard use cases.
Those shall faciliate decision support for practitioners and
researchers, willing to perform a big data-related project. The
creation of the intended solution, performed here, is equiva-
lent to the general design and development of the implicitly
followed design science research methodology.
A. SELECTION OF THE CLUSTERING ALGORITHM
To reduce the effort of analyzing the case studies, differ-
ent methods that are typically used for document cluster-
ing have been compared with each other in terms of their
applicability, as they are intensively investigated in various
contributions [78], [79]. In particular, partitioned, density-
based, and hierarchical types were investigated. While the
first intends to put similar objects to the same cluster while
maintaining the space between the different clusters as high
as possible [80], density-based approaches follow the idea
to identify regions where a high density (cluster) exists
and separate them from those regions with a lower density
(noise) [81], [82]. The last approach, on the other hand, is
the building of a tree structure where each node, except for
the leaf nodes, is a cluster that contains its children as sub-
clusters (dendrogram) [78], [79], [82], [83]. Although all of
the mentioned algorithms come with multiple potentials and
benefits, not all of them are applicable to the current problem.
Partitioned clustering, such as the K-means algorithm, for
instance, allows the use of multiple calculation methods for
cluster building [78]. However, some authors such as in [79]
critically observed this kind of algorithms, stating issues such
as the choice of the initial centroid, the strict predefined
number of clusters, and further highlight the importance of
other approaches [83].
A density-based approach, such as DBSCAN, tends to
work best with noisy data and outliers, but it is not
robust against high-dimensional data [83]. In this particular
approach, high-dimensional data must be processed and no
outliers are present. This results predominantly out of the
meticulous qualitative assessment of the contributions during
the literature review. Eventually, the hierarchical clustering
algorithm was chosen as a suitable alternative, avoiding the
need for starting parameters, specifying the strict number or
size of the clusters [81]. Furthermore, also high-dimensional
data can be recognized. Typically this approach structures
a given dataset and ‘‘provide[s] a view of the data at
different levels of abstraction’’ [79]. One of the most
VOLUME 8, 2020 186603
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
frequently used approaches is constituted by the agglomer-
ative clustering that assigns each object to one cluster and
merges them until a whole tree is formed. The first step
requires the calculation of a proximity matrix between the
objects. Following that, the two closest clusters, with the low-
est distance, are merged and the proximity matrix is updated
for the new cluster. The procedure is repeated until only one
cluster remains [79], [81], [83]. Compared to the K-means
and the DBSCAN algorithm, the hierarchical clustering does
not divide the points into end clusters, instead, a hierarchical
structure based on the unit distance is provided. In terms of
this, different results can be obtained through the alteration of
the used distance metrics and linkage functions [81]. Due to
the reason that all elements are connected to each other, out-
liers cannot be efficiently handled, but this was not required
for the given task. However, at the same time, qualitative
assessments can be realized through the manual definition of
a desired distance. In summary, the basic steps that are needed
for the clustering are the definition of the feature set for all use
cases, the creation of the input matrix, the examination of the
hierarchical clustering, the definition of the cluster structure
as well as the determination of the intercluster distance. After
everything is defined, the clusters need to be reviewed and
modified in case they are not correctly assigned. By the end,
those will be defined as standard use cases and thoroughly
described. To summarize all of the aforementioned steps
described before, an overview is depicted in Figure 2.
FIGURE 2. The use case analysis as a BPMN model.
B. CLUSTER ANALYSIS
As previously described, the standard use cases shall be
deduced from the found out clusters of the hierarchical clus-
tering, especially from a higher level of abstraction. In the
beginning, an input matrix needs to be defined. Initially,
the cases were collected and qualitatively checked by using a
modified version of the NIST template, describing the current
situation (e.g. represented by the aim and data characteristics)
as well as the obtained solution (used methods and tech-
nologies). Although the template formed a promising starting
point for the description of the feature matrix, it was not
possible to use it as a direct input for the clustering algorithm.
This is not only due to some unnecessary descriptive fields,
such as title, author, or the rough description of the use
case, but also for needed information that can be manifoldly
expressed like the variety of the data or the used algorithms.
After an additional examination of the filled templates, a total
of 30 binary features were identified. For the construction of
the input matrix, it was required to check each use case on the
base of a formulated feature set. Since the number of use cases
did not differentiate too much, compared to the number of
attributes, this task was manually performed. An overview of
all features together with the respective numbers of the index
and the occurrences is depicted in Table 4.
As one may note, some of the listed features were
more frequently identified compared to the others. In a
descending order this includes: Dynamic Data, Data Fusion,
Unstructured Data, Heterogeneous Data, Statistical Calcula-
tions, Multiple Sources, Big Data Analysis, Real-time Data,
Hadoop, and Batch Processing. The complete mapping of
all features and the respective use cases is given in Table 5.
While one column represents one feature, each row stands
for one use case. All of the features are related to the relevant
data characteristics, used methods, fields of application, and
also applied technologies. In terms of the data characteristics,
for instance, it was needed to clarify whether the data is
coming from various types of sources and if it should be
shared between different users or applications during the
operation phase. Following that, the type of file system was
checked, in particular, if a distributed file system (e.g. HDFS),
wide-area file system (e.g. Lustre) or parallel file system
was used. If a particular use case (row) fulfilled one of the
formulated features (column), a filled dot (•) was noted,
whereas for not related features, an empty dot (◦) was used.
Since most of the required functionalities of the intended
algorithm are already included within Matlab, this compu-
tational framework was used for the cluster analysis. The
created table was transformed into a binary matrix, filled
with ones and zeros, transposed, and eventually used for
the input. Besides the actual data, only a few inputs were
required in Matlab. This includes the used distance mea-
sure and linkage function [84]. While the first computes
the distance between each pair of observations, the second
uses the calculated distance between the observations and
links them according to the type of function that is cho-
sen [81]. In particular, the Euclidian distance and Ward’s
186604 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
TABLE 4. A list of all features and their occurrences.
linkage function were used, resulting in the dendrogram
depicted in Figure 3.
On the x-axis, it shows each of the previously listed use
cases. The y-axis describes the distance between the various
use cases and aggregated clusters. While most of the existing
clustering methods require a strict number of clusters or the
elements contained in them, the used approach requires only
minimum and maximum values for both, ranging from two
to n (as the number of cases). During the investigation and
formulation of the features, huge differences between the
cases were noticed in parts, which directly influenced this
range selection. By having too many clusters that differ only
slightly from each other, a decrease in the usability for later
standard cases can be expected. This would diminish the
general idea of standard use cases, especially when it comes
to the classification of a planned project for a potential user.
Apart from the time-consuming planning steps needed in
beforehand, also detailed knowledge about specific features
would be required to make further distinctions. This, in turn,
would neglect the general sensibility and applicability of the
targeted outcome. For that reason, the inter-cluster distance
together with the number of agglomerated clusters was exam-
ined to understand at which point all cases were assigned.
As one can note in the depicted diagram, at a distance of
two, only six clusters consisting out of multiple cases are
built, whereas 31 cases remain as one separate cluster. At the
level of 3.5, only one case remained unassigned and in total
13 clusters were formed. A distance of 4 resulted in seven
distinct clusters, which comprise all of the 43 use cases. At a
level of 4.5, only five agglomerated clusters exist. By having
the aforementioned disadvantage of too few cases in mind,
the achieved seven clusters at a distance of 4 were chosen.
A further qualitative assessment and cross-checking of each
of these seven clusters, however, revealed the disadvantage of
the non-weighed Euclidian distance function.
C. EXAMINATION OF THE BUILT CLUSTERS
Due to the reason that only binary decisions on the feature
set are recognized, no in-depth information extraction and
connections were made yet. As one can note, some of the use
cases revealed a rather high inter-cluster distance. For those
reasons, and to obtain a better understanding of the similari-
ties of those cases, further examinations were required. First
and foremost this includes the overall aim of the use case and
the interplay of the features. In the following, each of those
and their specifics are described in detail. Table 6 provides an
overview of the automatically built clusters and their assigned
use cases at a distance of 4.
1) DESCRIPTION OF CLUSTER 1
The first automatically built cluster comprises the seven use
cases no. 2[41], 3[42], 5[38], 6[71], 8[48], 37 [61] and
39 [63]. One particularity of the cases located in this cluster
that was noticed at the very beginning of the examination and
comparison, was the aim to improve already existing analysis
processes. On one side this comprises the automation of cur-
rently manually performed actions, like the financial rumor
detection [71], the monitoring of patients at their homes,
instead of using the hospital capacities [63], or the analysis
of traffic sensors data to recognize congestions and traffic
patterns [48]. On the other side, the improvement of existing
processes can be defined as achieving better quality of the
performed analysis. This can be realized by using all of the
available information, like unstructured medical and genome
data on the way to personalized medicine [63], investigating
the meaning of social media data to improve crisis mapping
systems [41] or the broadening of the scope [61].
All of those cases make use of unstructured data, coming
from various sources in different formats. Especially if those
have to be realized in real-time, sophisticated approaches
are required. This is for example the case for the autom-
atized financial rumor detection, considering more than
300 trades per day [71], or the simulation of wind turbine
VOLUME 8, 2020 186605
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
TABLE 5. Resulting matrix showing highlighting the occurrence of a feature (F) in a specific use case (U).
186606 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
FIGURE 3. Dendrogram of the cluster analysis, showing the cluster distance (y-axis) and use case number (x-axis).
TABLE 6. Automatically built clusters of the hierarchical clustering.
configurations resulting in a data stream of 100MBytes per
hour [61]. Furthermore, dynamic and permanent (historic)
data are used in all of the cases. The same hold true to
the application of statistical methods, like market activity
statistics [71] or the possibility of a medical condition to
occur in different situations [42], [63]. Moreover, the search,
query, and indexing of the data, used in the cases, aiming to
improve existing processes, should be enabled. The searching
process of the available data is an important part of the
analysis. To make the analysis of big data easier, the data can
be classified in different categories.
The consensus between all the cases is that they rely on
deep learning techniques to reach their goals. Those tech-
niques are for example needed to map financial news to
a trained set to decide the authenticity of the news [71].
In the medical sphere, machine learning is used to uncover
similarities between patients and thus to faciliate the proper
therapy prediction, or to enable the remote patient monitoring
by making sense of the medical records and pre-defined
rules [42], [63]. In [61] deep learning is used to analyze
different configurations of wind turbines, in order to decide
about their optimal location and design.
The technologies used on the way for the optimization
of existing processes differ from one another, as all of
those cases have their specific subtasks. Some of the cases
make use of the HDFS to deal with the volume of the
data [38], [48], [63], [71]. Those analyze some of the data
in batch mode to create a trained set, which is later used as
a basis for the real-time analysis and the decision-making.
Furthermore, in one of the cases studies, for the purpose of
reducing the dimensionality of the data, the parallel file sys-
tem Lustre is used [61], allowing the simulation of different
turbine configurations. For the implementation of the crisis
and mapping system [41], the ElasticSearch database is used,
as this one enables the near real-time processing of the data.
To enable remote patient monitoring and sensemaking of all
the collected sensor data, in [42], the analytics engine Spark
is used to speed up the query performance.
2) DESCRIPTION OF CLUSTER 2
The second cluster comprises the three use cases no. 1[36],
9[49], and 13 [43]. Each of them focuses on value creation
through the analysis of IoT sensor data. The data itself origi-
nates in each case from various sources. In [36], for instance,
smart meters, different appliances, and smart home devices
are used. Diverse weather sensors that measure temperature,
wind speed, and humidity are sourced in [49]. In the case
of [43], gas or imaging sensors that can typically be found in
recycling systems are focused. Apart from the behavior of the
occupants of smart buildings [36], or anomalies and failure
VOLUME 8, 2020 186607
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
detection of the weather sensors [49], also statistics over
the recycled goods and their usage to improve the recycling
system are targeted [43].
In each of the cases, different kinds of sensors are used.
Big data technologies are needed in all of them to cope
with the unstructured, large amounts of data. In particular,
comprehensive analysis and clustering algorithms are applied
to uncover yet unknown patterns, as they were described
before. For this reason, special pre-processing measurements
like data cleaning, which removes outliers and irrelevant
data, and measurements to structure the data have to be
applied [36], [49]. To receive the desired insights, detecting
the previously hidden patterns, a real-time data processing
is not necessarily needed. For example, for the detection of
failures in the weather sensors in [49], the data is firstly
collected and bulk-loaded into the IoT framework. Afterward,
the data is analyzed to identify similar values and potential
failures. The analysis of the data in the recycling system
is realized after the data from the different sensors were
collected and pre-processed [43]. The main aim of all use
cases of this cluster is to uncover some relevant patterns out of
the huge amount of IoT data. To reach this target, the gathered
data needs to be classified and subsequently analyzed with the
help of suitable algorithms. In the case of [36], the K-means
algorithm is used to discover the hourly usage of appliances
and their usage on different weekdays by the occupants of a
building. This algorithm is also used to discover patterns out
of the values, delivered from different weather sensors [49].
Another important part of the analysis process is the usage
of basic statistics. When the analysis is performed, the visu-
alization of the results is mandatory. For instance, this is used
to represent the hourly usage of different household devices
from the occupants of a building [36] or to show clusters,
built on the basis of the different weather sensor values [49].
As this group tries to realize the concept of IoT in different
areas of life, the sharing of information between users or
devices should be enabled. For example, the occupant of a
smart home should be able to exchange information with
his household devices like dishwasher or oven [36]. Another
example of such information exchange can be observed in
the smart recycling system, where retailers and consumers
should be also able to communicate [43]. The various uncov-
ered patterns can be used, for instance, to generate energy
reduction recommendations, to avoid failures in weather sen-
sors, and to increase the efficiency of the current recycling
systems [36], [43], [49].
3) DESCRIPTION OF CLUSTER 3
Based on the set of given features and their occurrences in the
respective use cases, no. 17 [54], 19 [44], 21 [66], 22 [67],
25 [58] and 30 [59] were automatically assigned to the third
cluster. Again, a thorough investigation and comparison was
made to identify conspicuous similarities. Three of the cases
are aiming to realize smart city concepts [44], [59], [66]. The
strong connection of those can solely be observed accord-
ing to the low inter-cluster distance in Figure 3, but also
regarding their overall scope. For example, in [66] the concept
of an itinerary planning platform for tourists, which suggests
activities according to pre-defined criteria, such as location,
time, and period preferences is proposed. In the case of [44],
unstructured data is used to improve transportation. In doing
so, information about incidents on the highway shared by
tweets on Twitter, videos of a disaster, or pictures of a
traffic jam are utilized. A touristic recommender system is
shown in [59]. The personalized recommendations are not
only based on permanent data, such as information about
city infrastructure, existing restaurants, or hotels, but also
dynamic data that is constantly changing. The latter is not
only related to the velocity of the data but also its structure.
These data include, for instance, information coming from
wearable bracelets, social networks, and used sensor data
applied for the traffic and weather tracking [44], [59], [66].
Eventually, all of those cases consider data coming from
different sources. Hence, the data fusion plays a dominant
role in all of those cases.
In this context, geographic information systems (GIS) are
used to gather, store and analyze the whole geographic infor-
mation such as the location of users, traffic jams, incidents,
disasters, hotels, restaurants or attractions [44], [59], [66].
For the actual provisioning of personalized recommenda-
tions, real-time data analyzes are required. In [44] those are
realized through the use of Spark streaming. Furthermore,
in each of those, sophisticated statistical methods are applied.
As the steps of data gathering, processing, and analysis are
conducted, the results need to be represented understandably
and appealingly for the user. Therefore, the usage of various
visualization techniques is crucial for this group. For instance,
the personal touristic recommendations for activities, accom-
modations, and restaurants in [59] are shown. Another exam-
ple of visual techniques is the representation of the current
transportation situation for a selected region in video or image
format [44].
However, this cluster contains three further cases that have
not been considered yet [54], [58], [67]. One of them rep-
resents a system for the remote 24/7 patient monitoring,
which can be also used to determine the future medication
procedure [54]. The second one represents a framework that
utilizes the data originating from the financial sector and
various IoT devices to improve the user experience [58].
Although connections to the previously described cluster
were identified, distinctions in terms of the data processing
and the format were observed. In particular, real-time and
batch processing on differently structured data is performed.
The last case study makes use of high-performance comput-
ing (HPC) to analyze and optimize the operation of wind
turbines [67]. This one also deviates from the smart city
concept, represented in the first three cases. Despite the fact
that those are serving in here as outliners, all of them share
other characteristics, such as the processing speed, used data
formats, and the overall aim to improve existing methods.
Due to those similarities, they could serve as a cluster or
rather a group themselves.
186608 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
4) DESCRIPTION OF CLUSTER 4
Cluster four contains the cases no. 4[47], 14 [51], 16 [53],
23 [56], 33 [70], 34 [75] and 35 [60]. During the observation
and further examination of each of those, a similar distinc-
tion as in the previous cluster was observed. In particular,
two sub-groups were identified, since they do not share one
sub-cluster but instead multiple characteristics.
The first sub-group comprises the cases originating
from [47], [56], [60]. They have in common that the main
focus is put on the realization of smart city concepts. The
first case in this group attempts to integrate information from
sensors and IoT devices used in a building, weather informa-
tion sensors as well as data from environmental sensors in
a cognitive building framework [47]. This framework shall
improve energy consumption by learning from the behavior
of the habitant and adjusting the functionality of the devices
accordingly to the users’ behavior [47]. The second case
introduces a smart traffic pilot, making use of the traffic light
data, weather and disaster information, as well as GPS data
about the positions of the vehicles. This information can be
used in different applications like route optimization or a
driving coach, suggesting fuel-saving driving patterns [60].
The last case in this group proposes a platform that integrates
data from IoT devices, GIS, and energy-related information
to improve the energy consumption and to reduce the CO2
emissions [56]. Each of those three concepts corresponds to
the general aim of realizing the smart city concept.
Notwithstanding that, all of them make use of data, coming
from various devices like household appliances, environmen-
tal sensors, smart meters, or GPS devices [47], [56], [60]. This
heterogeneous data is mostly unstructured and can have dif-
ferent content formats, in the form of texts, images, or videos.
Furthermore, personally identifiable information about the
habits of the buildings’ occupants, GPS, and passengers in
a car are used [47], [56], [60]. The personally identifiable
information together with the input about the city and build-
ings’ infrastructure make up the permanent data used in this
group. Moreover, the real-time processing or at least the near-
real-time processing of the data should be enabled by the
proposed smart city concept. For example, GPS data and
environmental information should be processed on the fly so
that a plausible route recommendation can be delivered [60].
Further, the near-real-time integration of data, coming from
IoT devices and various networks (e.g. electrical and heating
networks), is a requirement for the energy management sys-
tems, proposed in the last case in this group [56].
Additionally to those aspects, it was found out that all cases
use GISs to locate the user, particular vehicles, or relevant
objects. To represent the delivered smart city solution, the
cases deploy different visualization techniques. For example,
the analyzed driving behavior of the user and the derived fuel
economy recommendations can be represented in a mobile
application [60]. This solution was also used in [56] to visu-
alize the energy consumption data and the possible improve-
ments that can be conducted. To derive the above-mentioned
recommendations, various statistical methods are used.
Those methods are needed to calculate values such as the
energy consumption in particular rooms, average fuel con-
sumption on a road segment, or corresponding indicators for
certain time frames [47], [60].
The remaining cases are no. 14 [51], 16 [53], 33 [70] and
34 [75]. Those have different application areas and neither fit
in the smart city concept nor can build up a separate group.
The first one presents a smart clinical workflow implementa-
tion that should automate some parts of the patient care [51].
This one could fit to the smart city concept, but the developed
solution does neither make use of visualization techniques
nor a GIS and thus does not fit in the already built subgroup.
The second one [70] differs strongly regarding the aim to
integrate several bioinformatics databases and, thus, improve
the scalability of the cancer analytical system. The next case
study has a similar aim to the previous one. Here, the query
performance of a library information system needed to be
improved [53]. However, it has a different pattern of the ful-
filled feature set, including only structured data from only one
source. The last case study considers event-manufacturing
data and aims to improve existing processes [75]. This one
does not maintain personally identifiable information and
does not use GIS, which is a crucial requirement of the above-
defined smart city subgroup.
5) DESCRIPTION OF CLUSTER 5
The next cluster, derived from the hierarchical clustering
results, consists of three cases, namely no. 15 [52], 28 [74]
and 29 [46]. The general aim of those cases can be described
as the integration of data from different sources, improving
the scalability and leading to better analysis results. For
example, the first case study proposes a system for real-time
traffic control [52]. In comparison to the existing traffic con-
trol systems, the proposed solution should be able to consider
more than one aspect by involving more data resources. The
second case study aims to improve the quality of the user
experience in the communications area by involving more
resources in the analysis. In the past, data mining methods
were used in the telecom area to figure out problems only
in an isolated way, for example fraud detection based on
call detail records. In order to consider different telecom-
munication aspects, nowadays various information, coming
from mobile networks, GPS devices and social media has to
be considered [74]. The target of the last case study in this
category is to turn regular factories into smart factories, where
resources and machines communicate and deliver smart prod-
ucts that are aware of their production history. To realize those
factories, the integration of data from different machines and
operators needs to be pursued [46].
The data used in this cluster can be both, structured and
unstructured. Examples for structured data can be records
of the log and machine operating times [46], [74]. In con-
trast, social media data, camera pictures, and sensor data
are examples of unstructured data [46], [74]. Furthermore,
all three case studies maintain personally identifiable infor-
mation, which can require special processing techniques.
VOLUME 8, 2020 186609
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
This information can be the location of a driver, smart card
data of public transport users, or the call history log of a
telecom company customer [52], [74]. Permanent as well as
transient data (data that is deleted by the end of a session) play
an important role in the performed analysis in this cluster.
Regarding the processing of the data, real-time analytics
are required. For instance, to enable real-time traffic con-
trol, incident detection has to be performed on the fly [52].
The same applies to the communication between prod-
ucts, resources, and machines to realize smart factory con-
cepts [46]. Besides that, in all of the cases, NoSQL databases
are used to enable the querying of the diverse information.
In the presented solutions, basic statistics are used to calculate
values such as average speed, travel time, subscriber churn
likelihood, and operational time of a machine [46], [52], [74].
Deep learning algorithms are used to reach the goals of the
analysis. Those enable, for example, the consideration of
user ufeedback in the analysis of the users experiences’s
quality [74].
6) DESCRIPTION OF CLUSTER 6
The sixth cluster includes the seven case no. 7[64], 10 [37],
11 [65], 26 [73], 27 [45], 42 [40] and 43 [33]. When inspect-
ing the cases, two different groups were found out.
The first one consists of the four use cases no. 7[64],
27 [45], 42 [40], and 43 [33], aiming to deal with the growing
amount of medical data and to improve the analysis quality in
the healthcare area through the integration of additional data.
In the first use case of this group [64], a Hadoop ecosystem to
deal with the volume, variety, and velocity of medical data,
coming from various applications and devices is proposed.
The use case no. 42 [40] deals with the analysis of human
genome data, which is continuously growing and even HPC
clusters cannot deal with the challenge to process this amount
of data. As a solution to the problem, a Hadoop system is
proposed. In [45] a framework that allows querying both
structured and unstructured medical data is introduced. The
main goal is to improve the decision basis for medical experts.
The last case [33] in this group presents a cloud-based
analytics solution that should turn the massive amounts of
medical data into value. In almost all cases, structured and
unstructured data, located in the medical area, are used. While
the first mainly contains structured documents like patient
records like the name, age, or previous diagnosis, the latter
comprises images, clinical notes, unstructured documents,
and genome data [33], [40], [45], [64]. Due to the handling of
personally identifiable information, which falls under special
regulations, sophisticated security measurements and storage
solutions are required in the final system [33]. Again it was
noticed that data fusion is of major importance, because of
the use of data originating from different (healthcare) insti-
tutions, devices, and other sources. For example, in [64] a
system is proposed that is intended to improve the healthcare
situation in Algeria by an efficient distribution of medical
resources and staff. To achieve this, information from one
university hospital, five public hospitals, one medical school,
51 polyclinics and some laboratories needs to be integrated.
For the technical implementation, all of the use cases rely on
the HDFS. Furthermore, it was found out that in none of the
use cases a real-time processing was needed [40], [64]. For
the analysis itself, all use cases utilize basic statistics and data
mining.
Compared to the cases described before, all of the remain-
ing, including 10 [37], 11 [65], and 26 [73], follow a differ-
ent aim. In here, the linkage of data from different sources
is of major interest. The first case [37] of those presents
a framework for the detection of insurance frauds. At the
moment, insurance fraud detection methods are solely being
used in separate fields like healthcare, financial services, and
others. To find a suitable solution and to achieve a broad cov-
erage, data from 34 different fields, including sources such
as customer information, contracts and insurance claims, are
integrated [37]. Case no. 11 [65] presents an architecture for
the processing of data coming from different social media
channeld and in [73] a data management system environment
that is supposed to deal with unstructured data from various
sources is introduced. All of those case studies require the
processing of unstructured data [37], [65], [73]. Most of
the considered data such as insurance contracts and claims
are in an unstructured textual form [37]. As the cases in
this subgroup propose frameworks and architectures for the
integration of the vast amount of data from different sources,
the cleaning of the original data constituted an important
step. This includes auxiliary activities, like outlier detection
or fixing missing values [37], [73]. Similar to the previous
group, real-time data processing is not required.
7) DESCRIPTION OF CLUSTER 7
The last cluster, resulting from the hierarchical clustering,
contains the ten use cases no. 12 [50], 18 [55], 20 [72],
24 [57], 31 [68], 32 [69], 36 [76], 38 [62], 40 [77] and
41 [39]. Similar to the sixth cluster, the use cases can be split
in two separate groups, regarding their reasons for using big
data technologies.
The first group consists of the three cases [55], [62], [68]
that are focused on supporting informed decision making,
mainly in the healthcare area. The first case study attempts
to provide a basis for precision medicine, by facilitating the
data analysis of various molecular profiles [55]. The second
proposes a general framework that should enable the integra-
tion of healthcare data from various resources and thus allow
researchers to conduct innovative types of analysis [62]. The
last case in this group aims to improve the daily practices
in a hospital by utilizing transactional data [84]. All of the
three cases make use of heterogeneous, unstructured data of
different content formats, such as text, images (e.g. diagnostic
tests), signals, and phenotypes [62]. Besides that, personally
identifiable information, in the form of patient records, are
used [55], [62], [68]. A further important feature that is shared
in all of the case studies, is the initial data pre-processing.
Here, it constitutes a crucial step as it increases the quality
of the information used for the decision-making process. It is
186610 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
for example used to deal with spelling and grammar mistakes
or anonymization [55], [62], [68]. The analysis, used for the
decision making, is performed by the help of basic statistics
and data mining techniques. Different associations, classi-
fication, and clustering methods are incorporated, in order
to discover relevant patterns, as well as to figure out inter-
attribute correlations and important relations [62], [68].
The second group contains the remaining six documents,
namely the cases no. 12 [50], 20 [72], 24 [57], 32 [69],
36 [76] and 41 [39]. All of them share the aim to enable
the real-time analysis of data, incoming with high-speed.
In the first of the case studies, a framework for the online
analysis of high-speed physiological data is proposed, which
should improve the neonatal intensive care [50]. The sec-
ond one uses real-time analysis to forecast the power output
of solar plants [72]. Within the third case study, a frame-
work for the analysis of patent information and its usage
for research and development is introduced [57]. The sub-
sequent case no. 32 [69] proposes an analytic platform for
smart transportation, which analyzes data from heteroge-
neous data sources such as sensors and cameras in real-time.
In [76] real-time analysis is used to process social media
data for a disaster management system. The last use case
uses social media data to improve the quality of passenger’s
experience [39]. As one may note, all of the described use
cases in this group focus on real-time data analysis. The
origin for this circumstance arises presumably out of the high
velocity.
A further important feature of this group is the processing
of heterogeneous data that is coming from different devices
or source. For example, cameras and traffic sensors are used
as an input for the realization of a smart transportation system
in [69]. Apart from that, social media data, for example
provided by Twitter, Google+, or YouTube, can be harnessed
for disaster management or the improvement of the quality
of the customer’s experience [39], [76]. For case no. [72]
structured, semi-structured (e.g. weather forecast data) and
unstructured data (e.g. customer behavior, video files) are
used. Because of the flexibility and scalability, the HDFS is
incorperated as the foundation for each use case. The final
results are eventually visualized and presented for instance
by dashboards [72], bar charts [57], or a decision map [76].
D. DERIVED STANDARD USE CASES
From the first explanation and qualitative examination of the
seven found out clusters it was noted that some of the use
cases did not match properly to each other, even though they
were assigned to one agglomerated cluster. Hence, modifi-
cations were required in different ways, such as insertion,
deletion, and consolidation, to better highlight the data char-
acteristics, used methods, and aim of the use case. Above
all, this was required to ensure that not only key indicators,
such as the distance, are used for the creation of the stan-
dard use cases, but also qualitative assessments are realized.
Subsequentelly, in total, nine different clusters were derived
from the qualitative examination. Those are depicted
in Table 7.
In the first and second column, the number of the derived
cluster as well as the mapped use cases are stated. The
general aim, all relevant features, and the needed modifica-
tions are described in the remaining columns.For instance,
it was noted that the first identified groups of clusters three
and four, containing cases no. 19, 21, 30, and 4, 23, 35,
share similar interests. Besides the general focus on smart
cities, also the same characteristics are shared, except for
one case (no. 23) that uses near-real-time processing instead
of real-time-processing [56]. Hence, both of them were
merged into one new cluster (cf. Table 7 cluster no. 3).
All remaining cases of the initially calculated third clus-
ter, focusing on sensor analysis, became the new second
cluster.
Furthermore, the cases 16, 33, 34, 40 appeared to be as
outliers, not only for the respective fourth cluster but in parts
also for the entire dataset. This does not represent an error
in the qualitative analysis but rather how heterogeneous the
individual use cases can be. For instance, in case no. 16 [53],
the main goal was to improve the query performance of
a library information system. Within the given collection,
this was the only case that solely handled structured data.
Case no. 33 [70] exclusively used graphical-processing of
the data to efficiently handle queries on multiple integrated
bioinformatics databases. To prevent misleading information,
those use cases were removed or assigned to another cluster.
Case no. 14 [51] proposes a smart clinical workflow that aims
to increase the volume of medical data that can be processed.
In doing so, health data from different sources are integrated
and used to facilitate predictive therapy and to improve the
wellbeing of the patient. Due to the similarities to the first
cluster and the goal to generally improve the quality of the
performed analysis, this case was assigned to said cluster.
As already highlighted during the description of the orig-
inal third cluster, the cases no. 17 [54], 22 [67] and 25 [58]
revealed no real interconnection to the overall features pre-
sented in this cluster. By comparing those cases, it was found
out that all of them aim to optimize existing processes by
using big data. Eventually, a new cluster was manually built
(cf. Table 7 cluster no. 9). The separate groups, which were
identified within the initially calculated sixth and seventh
cluster were extracted and declared as a separate one. Hence,
out of both clusters, two additional ones emerged (cf. Table 7
cluster no. 5-8).
Overall, for most of the initially formed clusters, only
minor modifications were required. Researchers, as well as
practitioners, can utilize these to obtain an idea not only
about the general meaningfulness of their own project but
also possible implementation details from specific use case
descriptions. This is especially the case if a similar approach
is pursued. For an increased understandability, within the
following paragraphs, each of the found out standard use
cases are briefly described, comprising the common features
in a narrative way.
VOLUME 8, 2020 186611
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
TABLE 7. Standard Use Cases (UC) derived from the clusters. TABLE 7. (Continued.) Standard Use Cases (UC) derived from the clusters.
1) STANDARD USE CASE 1 – DATA ANALYSIS
IMPROVEMENT
By adopting big data technologies, an improvement in the
quality of the data analysis is pursued. A significant step
to achieve this aim is to make sense of massive amounts of
unstructured data that are coming with high-speed, and the
exploitation of sophisticated methods, such as deep learn-
ing. Additionally to that, statistics and classification methods
are often used to increase the quality of the analysis. The
described characteristics of this general case and the used
methods can be mapped to different cases, coming from
healthcare, transportation, manufacturing areas, and social
media. Details of the particular cases can be viewed in [38],
[41], [42], [48], [61], [63], [71], [85].
2) STANDARD USE CASE 2 – BATCH MODE SENSOR DATA
ANALYSIS
One of the reasons for harnessing big data technologies is
to enable the processing of large amounts of (IoT) sensor
data to obtain new insights. Key factors in this use case
are the integration of different data sources, such as sen-
sors and devices, as well as enabling the data exchange
between users and applications. The data commonly does
not exist in a structured format, thus processing unstructured
data plays an important role. Real-time processing is not
required, as the data is firstly gathered and then processed in
batch mode. To uncover different types of patterns, clustering
approaches are used for the analysis. The visualization of the
processed data is crucial to represent the findings. Based on
those, strategies, for instance, to improve the user experience,
resource allocation, process costs, and others, can be devel-
oped. Concrete specifications for this standard use case are
explained in [36], [43], [49].
3) STANDARD USE CASE 3 – SMART CITY
This category deals with the challenges of smart cities by
involving various resources in real-time data analysis. The
concept itself utilizes data from various devices, sensors,
and human actors to improve the quality of life for citizens.
For this purpose, structured, unstructured as well as transient
and permanent data can be used as analysis input. In order
to turn a large amount of heterogeneous data into value,
deep learning algorithms are used. In this case, a robust
storage solution for massive amounts of differently structured
186612 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
data should be used, such as a NoSQL database. Due to
the nature of this domain, personal information have to be
recognized and privacy-preserving techniques are applied.
All related cases are comprehensively described in [44],
[47], [56], [59], [60], [66].
4) STANDARD USE CASE 4 – MULTI-LEVEL PROBLEMS
In this standard use case, sophisticated multi-level problems
are stated, which require thorough planning from different
perspectives, covering not only the needed system but also the
data being processed. Organizations facing those problems
are confronted, in particular, with the growing amount of data
coming from various institutions, such as in the healthcare
sector. Apart from the required high reliability of the targeted
solution and the needed ability to efficiently search, query
and store the data, also privacy-preserving techniques have to
be considered. Moreover, processing unstructured data, such
as handwritten documents or images, needs to be enabled.
For the analysis of the data, different data mining approaches,
which analyze stored data (e.g. on an HDFS) in batch mode,
can be considered. This standard use case originates from the
following contributions [46], [52], [74].
5) STANDARD USE CASE 5 – EXPAND DATA SOURCING
In this case, data coming from various resources needs to
be combined into one functioning system. As the consid-
ered data originates from different sources or instances, not
only the structure but also the data itself can be highly
volatile. Due to this reason, not only sophisticated stor-
age solutions for those various types of data (e.g. NoSQL),
but also sophisticated pre-processing techniques are needed.
After the initial collection and cleaning, various statistical
methods can be used. The data is usally processed in batch-
mode. Concrete details of all relevant use cases can be found
in [33], [40], [45], [64].
6) STANDARD USE CASE 6 – DATA CONNECTION
Adopting big data technologies in areas with widespread
collections of information can improve decision-making
by incorporating a larger information basis. As wrong
decisions, especially in domains like healthcare, can have
enormous consequences, guaranteeing the correctness of the
analyzed data is a significant step, necessitating extensive
pre-processing. Depending on the application area, this can
additionally require special processing steps like anonymiza-
tion or classification. For the analysis, data mining techniques
can be used and also efficient querying and searching over the
data in real-time should be enabled. Further information are
provided in [37], [65], [73].
7) STANDARD USE CASE 7 – DECISION SUPPORT
Real-time analytics on differently structured data are used in
those use cases, to facilitate decision support for data-driven
problems. Through basic statistics, classifications and other
analytical methods, previously unused ata are converted into
valuable information. For a better presentation of the obtained
results, visualization techniques are highly important. This
use case can be characterized by the phrase turn volume into
value. Details can be observed in [55], [62], [68].
8) STANDARD USE CASE 8 – HIGH-SPEED ANALYSIS
Within this use case, the input data comes in a structured
and unstructured format and needs to be processed in (near-)
real-time, to ensure that all functionalities and results can
be immediately provided. In addition, to maintain, search,
query, index and analyze all data, complex solutions are
required. For a comprehensible representation of the results
and the performed calculations, visualization techniques are
paramount. For particular insights, the following contribu-
tions can be used [39], [50], [57], [69], [72], [76].
9) STANDARD USE CASE 9 – PROCESS OPTIMIZATION
Big data technologies turned out to be an enabler for the
general optimization of existing processes. Usually, the data
is incoming with high velocity and needs to be processed in
real-time. However, also batch-processing mode should be
available either as a backup solution or for specific analytical
tasks. In this case, both, structured and unstructured data
are considered. Clustering techniques support the identifica-
tion of recommendations with which existing processes can
be optimized. Various visualization techniques allow for a
better presentation in an appealing way. Further details are
described in [54], [58], [67].
IV. EVALUATION
In order to check the validity of the artifact and, thus, the pro-
posed standard use cases, a thorough evaluation is required
at which multiple aspects are verified [27], [28]. On the
one hand, it is necessary to assure a sufficient coverage of
the regarded application domains, methods as well as data
characteristics and on the other hand, the undertaken trade-off
between possible degrees of fragmentation has to be looked at
(cluster building). To assess the coverage with a practical ori-
entation, an approach that is inspired by machine learning’s
division into training and test data is utilized. For this purpose,
the steps of the literature review are replicated while applying
the same criteria (cf. Table 1), but this time only for selected
case studies published in 2019.
Apart from the search procedure, also the comprehen-
siveness check through the use of the altered template was
performed. In total three additional use cases were found and
used for the evaluation of the found out results. Since those
cases were not involved in the creation of the standard use
cases, they function as the equivalent of a test data set. The
first case study used for the evaluation comes from the area
of online retail [86]. It provides an approach for a recom-
mendation system that can be realized in an online store,
requiring a user to sign up. Besides harnessing historical
and transactional data, it also makes use of the customer’s
browsing history. Taking a look at Table 7 and considering
that the case study involves both structured as well as unstruc-
tured data and requires real-time processing, regarding those
VOLUME 8, 2020 186613
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
FIGURE 4. The DSR Grid according to (Vom Brocke and Maedche, 2019).
features, this one can be placed in the fourth, eighth, and ninth
cluster. However, the used recommender engine in this case
study has a key role in the data analysis, which reduces the
categorization possibilities only to the ninth cluster.
In conclusion, the analyzed case study aims to improve the
online retail by introducing personal recommendations, based
on real-time processed browsing data, fitting the ninth general
use case that has the target to optimize existing processes
with the deployment of big data technologies. The second
case study presents a system for the incorperation of real-time
social media data in the analysis in the area of tourism [87].
The analysis of the data comprises the following main steps
– gathering of the data, cleaning and storage, querying and
filtering as well as the visualization of the results. The data
is collected from different social media sources in this case
– Instagram, Flickr, Foursquare and Twitter, resulting in an
unstructured content format that can manifest in the form
of posts, reviews, images or videos. Utilizing Table 7 and
considering the case study’s aim to involve real-time social
media data in the analysis, this one can be placed in the eighth
cluster, which targets real-time analysis of data, incoming
with high-speed. The last case study, harnessed to evaluate the
identified use cases, originates from the area of smart trans-
portation [88]. Compared to the already existing approaches,
which deal with single issues like congestion avoidance or
environmental-friendly driving, the considered case study
shows a system that proposes a solution to multiple problems.
It aims to track vehicles, suggest optimal routes and realize a
smart parking concept, utilizing predominantly unstructured
data from various sources like sensors, cars or navigation
systems. With regard to Table 7, this example fits into the
third general use case.
In conclusion, the successful categorization of the three
evaluation case studies in one of the defined general use cases
suggests that adequate coverage was achieved. The degree of
fragmentation, in turn, is based on the intended application
scenario. While a more general approach might increase the
coverage even further, it offers no clear orientation in the
selection of potentially similar case studies. Vice versa, every
case as its own category would effectively negate the idea of a
categorization. For that reason, the current number constitutes
a trade-off that allows for a choice of relevant properties,
while still providing several example cases as a knowledge
base for the aspired endeavor.
V. CONCLUSION
In recent years, big data has been one of the most promi-
nent topics in the IT-sector. However, there is still a lot of
unawareness and uncertainty when it comes to the execution
of such projects, especially right at the beginning of the
planning phase. Hence, in the contribution at hand, an in-
depth investigation of successfully conducted projects was
performed, to provide future practitioners as well as other
researchers, inter alia, with decision support concerning the
realization of their potential big data projects. As a result
of a literature review, 43 cases published between 2015 and
2018 were identified. Those cover detailed information about
the presented big data projects. To achieve a categorization
for the obtained results, all use case descriptions were thor-
oughly examined using a textual analysis technique. At this
stage, the hierarchical clustering proofed to be a promising
solution, revealing various clusters with a similar feature set.
Based on the gathered information and further modification,
a total of nine distinct clusters were identified.
186614 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
TABLE 8. Amended and adapted use case template.
VOLUME 8, 2020 186615
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
TABLE 8. (Continued.) Amended and adapted use case template.
Subsequently, those standard use cases constitute the arti-
fact of the conducted DSR endeavor as well as the answer to
the research question. To summarize and highlight the main
pillars, implications and key aspects of this research, in the
following, the corresponding DSR grid according to vom
Brocke and Maedche [89] is depicted in Figure 4. One part
of the contribution is constituted by the collection, structur-
ing and presentation of comprehensively described use cases
published in recent years, in the academic area. Additionally,
a template was used and modified for the analysis of the
identified cases. Through the use of this template, the gen-
eral comprehensiveness of one’s endeavor can be validated
and possible shortcomings or unrecognized gaps identified.
Beyond that, a presentation of standard use cases, derived
from the investigated publications, is made that can serve as
an orientation and initial starting point for the realization of
related projects. Consequently, researchers as well as practi-
tioners may greatly benefit from the results discussed in this
work.
A. LIMITATIONS AND FUTURE RESEARCH
Although a suitable answer to the initially formulated RQ
was achieved, certain aspects have to be mentioned, which
may call for future optimization or new research directions.
This refers not only to the results as such but also to the
methods used to achieve them. For instance, this includes the
recognition of additional weightings during the analysis of
the input matrix, since sometimes a particular feature appears
to be more important than another one. An example for this
are features that are directly related to the data characteristics
or methods used to analyze them.
During the qualitative analysis of the use cases it was
noticed that many of the project descriptions also contained
concrete specifications. However, most of the decisions are
tailor-made. Hence, an additional investigation of them may,
in turn, result in an even more complicated analysis of the
data. For now, implementation details can be viewed in each
of the referred cases within the standard use cases. Addi-
tionally, the chosen algorithm represents only one suitable
way for the creation of the found out clusters. Apart from
the discussion of the available algorithms and their poten-
tial usability, also other algorithms were alternatively tested,
especially with a view on future enlargements of the dataset,
for which the manual processing of each case would require
too much effort. Hence, for the creation of the input matrix,
a computer-supported solution was tested that automatically
processes the data and identifies important phrases on the
base of the term frequency. Despite a thorough pre-processing
procedure, which focused on the cleaning of unnecessary stop
words, endings, and inconsistent descriptions, no promis-
ing results were found. Even after an additional filter-
ing step of the found out phrases, the clusters had too
many dissimilarities. This was not only assessed on the qual-
itative level but also due to multiple irrelevant phrases, such
186616 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
as ‘‘the authors’’ or ‘‘it has’’. Nevertheless, as already shown
before, this has not achieved the desired effect.
Another tested method was the use of natural language
processing to uncover existing topics in a collection of unpro-
cessed textual documents [90]. In particular, the topic mod-
eling approach LDA was examined, which is a probabilistic
model that considers each topic as a combination of keywords
and each document as a combination of multiple topics.
Even though comprehensive pre-processing steps, such as
lemmatization, stop words and punctuation removal, were
repeatedly performed [90], no satisfying results were found.
Frequently, it was noticed that buzzwords, especially used
in the introduction and conclusion of the papers, have been
identified.
The evaluation of the coverage of the formulated cases
had a positive result, however, the sample size was rather
small and especially future big data projects might poten-
tially necessitate adjustments. Consequently, the sample size
of the found big data projects could be enlarged. At this
point, an extension of the actual dataset could be realized
through the investigation of additional years, further literature
databases and also by interviewing larger companies, that
conduct big data-related projects. Beyond that, also a long-
term evaluation is planned, which shall be realized through
the application at the very beginning of a project. Here, not
only the general meaningfulness but also the possibility to
derive concrete information and implications from the indi-
vidual use cases could be tested.
By referring to this, an implementation of the derived stan-
dard use cases within a concrete decision support system for
big data projects, for instance conceptually described in [23],
appears to be a promising direction for future research.
APPENDIX
See Table 8.
REFERENCES
[1] C. Dobre and F. Xhafa, ‘‘Intelligent services for big data science,’’ Future
Gener. Comput. Syst., vol. 37, pp. 267–281, Jul. 2014, doi: 10.1016/j.
future.2013.07.014.
[2] S. Yin and O. Kaynak, ‘‘Big data for modern industry: Challenges
and trends [point of view],’’ Proc. IEEE, vol. 103, no. 2, pp. 143–146,
Feb. 2015, doi: 10.1109/JPROC.2015.2388958.
[3] X. Jin, B. W. Wah, X. Cheng, and Y. Wang, ‘‘Significance and challenges
of big data research,’’ Big Data Res., vol. 2, no. 2, pp. 59–64, Jun. 2015,
doi: 10.1016/j.bdr.2015.01.006.
[4] L. Zhu, F. R. Yu, Y. Wang, B. Ning, and T. Tang, ‘‘Big data analytics in
intelligent transportation systems: A survey,’’ IEEE Trans. Intell. Transp.
Syst., vol. 20, no. 1, pp. 383–398, Jan. 2019, doi: 10.1109/TITS.2018.
2815678.
[5] W. L. Chang and N. Grady, NIST Big Data Interoperability Framework—
Definitions. Gaithersburg, MD, USA: NIST, 2019. Accessed: Jul. 14, 2020,
doi: 10.6028/NIST.SP.1500-1r2
[6] A. Gandomi and M. Haider, ‘‘Beyond the hype: Big data concepts, meth-
ods, and analytics,’’ Int. J. Inf. Manage., vol. 35, no. 2, pp. 137–144,
Apr. 2015, doi: 10.1016/j.ijinfomgt.2014.10.007.
[7] S. Kaisler, F. Armour, J. A. Espinosa, and W. Money, ‘‘Big data: Issues
and challenges moving forward,’’ in Proc. 46th Hawaii Int. Conf. Syst. Sci.,
Jan. 2013, pp. 995–1004.
[8] D. Izadi, J. Abawajy, S. Ghanavati, and T. Herawan, ‘‘A data fusion method
in wireless sensor networks,’’ Sensors, vol. 15, no. 2, pp. 2964–2979,
Jan. 2015, doi: 10.3390/s150202964.
[9] H. Lee, N. Aydin, Y. Choi, S. Lekhavat, and Z. Irani, ‘‘A decision sup-
port system for vessel speed decision in maritime logistics using weather
archive big data,’’ Comput. Oper. Res., vol. 98, pp. 330–342, Oct. 2018,
doi: 10.1016/j.cor.2017.06.005.
[10] A. P. Plageras, K. Psannis, C. Stergiou, H. Wang, and B. B. Gupta. (2018).
Efficient IoT-Based Sensor BIG Data Collection-Processing and Analysis
in Smart Buildings. [Online]. Available: https://www.semanticscholar.
org/paper/Efficient-IoT-based-sensor-BIG-Data-and-analysis-in-
Plageras-Psannis/fb18e87bdfa27b3bc7a9d9337f02cd6b66d0c372
[11] K. E. Psannis, C. Stergiou, and B. B. Gupta, ‘‘Advanced media-based smart
big data on intelligent cloud systems,’’ IEEE Trans. Sustain. Comput.,
vol. 4, no. 1, pp. 77–87, Jan. 2019, doi: 10.1109/TSUSC.2018.2817043.
[12] Y. Wang, L. Kung, W. Y. C. Wang, and C. Cegielski, ‘‘Developing a big
data-enabled transformation model in healthcare: A practice based view,’’
in Proc. 25th Int. Conf. Inf. Syst., 2014, pp. 1–12.
[13] P. Aversa, L. Cabantous, and S. Haefliger, ‘‘When decision support sys-
tems fail: Insights for strategic information systems from formula 1,’’
J. Strategic Inf.Syst., vol. 27, no. 3, pp. 221–236, Sep. 2018, doi: 10.1016/j.
jsis.2018.03.002.
[14] R. Häusler, D. Staegemann, M. Volk, S. Bosse, C. Bekel, and K. Turowski,
‘‘Generating content-compliant training data in big data education,’’ in
Proc. 12th Int. Conf. Comput. Supported Edu., 2020, pp. 104–110.
[15] T. Nguyen, L. Zhou, V. Spiegler, P. Ieromonachou, and Y. Lin, ‘‘Big
data analytics in supply chain management: A state-of-the-art literature
review,’’ Comput. Oper. Res., vol. 98, pp. 254–264, Oct. 2018, doi: 10.
1016/j.cor.2017.07.004.
[16] O. Müller, M. Fay, and J. vom Brocke, ‘‘The effect of big data and analytics
on firm performance: An econometric analysis considering industry char-
acteristics,’’ J. Manage. Inf. Syst., vol. 35, no. 2, pp. 488–509, Apr. 2018,
doi: 10.1080/07421222.2018.1451955.
[17] S. F. Wamba, S. Akter, A. Edwards, G. Chopin, and D. Gnanzou, ‘‘How
‘big data’ can make big impact: Findings from a systematic review and
a longitudinal case study,’’ Int. J. Prod. Econ., vol. 165, pp. 234–246,
Jul. 2015, doi: 10.1016/j.ijpe.2014.12.031.
[18] Z. A. Al-Sai, R. Abdullah, and M. H. Husin, ‘‘Critical success fac-
tors for big data: A systematic literature review,’’ IEEE Access, vol. 8,
pp. 118940–118956, 2020, doi: 10.1109/ACCESS.2020.3005461.
[19] D. Staegemann, M. Volk, A. Nahhas, M. Abdallah, and K. Turowski,
‘‘Exploring the specificities and challenges of testing big data systems,’’ in
Proc. 15th Int. Conf. Signal-Image Technol. Internet-Based Syst. (SITIS),
Nov. 2019, pp. 289–295.
[20] D. Staegemann, M. Volk, N. Jamous, and K. Turowski, ‘‘Understanding
issues in big data applications—A multidimensional endeavor,’’ in Proc.
25th Amer. Conf. Inf. Syst., 2019, pp. 1–10.
[21] S. Sagiroglu and D. Sinanc, ‘‘Big data: A review,’’ in Proc. Int. Conf.
Collaboration Technol. Syst. (CTS), May 2013, pp. 42–47.
[22] S. Bonesso, E. Bruni, and F. Gerli, ‘‘How big data creates new job
opportunities: Skill profiles of emerging professional roles,’’ in Behavioral
Competencies of Digital Professionals: Understanding the Role of Emo-
tional Intelligence, S. Bonesso, E. Bruni, F.Gerli, Eds. Cham, Switzerland:
Palgrave Macmillan, 2020, pp. 21–39.
[23] M. Volk, D. Staegemann, M. Pohl, and K. Turowski, ‘‘Challenging big data
engineering: Positioning of current and future development,’’ in Proc. 4th
Int. Conf. Internet Things, Big Data Secur., 2019, pp. 351–358.
[24] M. Volk, D. Staegemann, S. Bosse, A. Nahhas, and K. Turowski, ‘‘Towards
a decision support system for big data projects,’’ in WI2020 Zentrale
Tracks, N. Gronau, M. Heine, K. Poustcchi, and H. Krasnova, Eds. Berlin,
Germany: GITO Verlag, 2019, pp. 357–368.
[25] R. Dontha. The Origins of Big Data—KDnuggets. Accessed: Jun. 16, 2020.
[Online]. Available: https://www.kdnuggets.com/2017/02/origins-big-
data.html
[26] O. Ylijoki and J. Porras, ‘‘Conceptualizing big data: Analysis of case
studies,’’ Intell. Syst. Accounting, Finance Manage., vol. 23, no. 4,
pp. 295–310, Oct. 2016, doi: 10.1002/isaf.1393.
[27] R. H. Von Alan, S. T. March, J. Park, and S. Ram, ‘‘Design science in
information systems research,’’ MIS Quart., vol. 28, no. 1, pp. 75–105,
2004.
[28] K. Peffers, T. Tuunanen, M. A. Rothenberger, and S. Chatterjee, ‘‘A design
science research methodology for information systems research,’’ J. Man-
age. Inf. Syst., vol. 24, no. 3, pp. 45–77, Dec. 2007.
[29] J. Webster and R. T. Watson, ‘‘Analyzing the Past to Prepare for the Future:
Writing a Literature Review,’’ MIS Quart., vol. 26, no. 2, pp. 13–23, 2002.
[Online]. Available: http://www.jstor.org/stable/4132319
VOLUME 8, 2020 186617
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
[30] Y. Levy and T. J. Ellis, ‘‘A systems approach to conduct an effective
literature review in support of information systems research,’’ Informing
Sci., Int. J. Emerg. Transdiscipline, vol. 9, pp. 181–212, Jan. 2006, doi:
10.28945/479.
[31] R. Bauer. 3 Reasons Why You Need Business Case Studies—PAN Com-
munications. Accessed: Mar. 18, 2020. [Online]. Available: https://www.
pancommunications.com/blog/3-reasons-why-you-need-business-case-
studies/
[32] R. K. Yin, Case Study Research and Applications: Design and Methods.
Los Angeles, CA, USA: SAGE, 2018.
[33] F. Khalid. (2017). Inovalation is Driving Healthcare Transformation With
Pre-Engineered Infrastructure and Big Data Analytics. Dell EMC, Bowie,
MD, USA. Accessed: Jun. 1, 2020. [Online]. Available: https://www.emc.
com/collateral/customer-profiles/inovalon-vscale-case-study.pdf
[34] W. L. Chang and G. Fox. (2018). NIST Big Data Interoperability
Framework—Use Cases and General Requirements. Gaithersburg, MD,
USA. Accessed: Jul. 14, 2020. [Online]. Available: https://nvlpubs.nist.
gov/nistpubs/SpecialPublications/NIST.SP.1500-3r1.pdf
[35] J. Vom Brocke, A. Simons, B. Niehaves, K. Reimer, R. Plattfaut, and
A. Cleven, ‘‘Reconstructing the giant: On the importance of rigour in
documenting the literature search process,’’ in Proc. ECIS, Verona, Italy,
2009, pp. 1–14.
[36] A. Yassine, S. Singh, M. S. Hossain, and G. Muhammad, ‘‘IoT big data
analytics for smart homes with fog and cloud computing,’’ Future Gener.
Comput. Syst., vol. 91, pp. 563–573, Feb. 2019, doi: 10.1016/j.future.
2018.08.040.
[37] D. Kenyon and J. H. P. Eloff, ‘‘Big data science for predicting insur-
ance claims fraud,’’ in Proc. Inf. Secur. South Afr. (ISSA), Aug. 2017,
pp. 40–47.
[38] Y. Zhang, M. Zhang, T. Wo, X. Lin, R. Yang, and J. Xu, ‘‘A scal-
able lnternet-of-Vehicles service over joint clouds,’’ in Proc. IEEE Symp.
Service-Oriented Syst. Eng. (SOSE), Mar. 2018, pp. 210–215.
[39] H.-M. Chen, R. Schütz, R. Kazman, and F. Matthes, ‘‘How Lufthansa cap-
italized on big data for business model renovation,’’ MIS Quart. Executive,
vol. 16, no. 1, p. 4, 2017.
[40] M. Schlesner and F. Schinkel. (2016). Big Data Use Case: Genomic Data
Research. Fujitsu, Munich, Germany. Accessed: Sep. 4, 2019. [Online].
Available: https://www.datameer.com/wp-content/uploads/pdf/misc/cs-
PF4H-Genome-Research.pdf
[41] M. Avvenuti, S. Cresci, F. Del Vigna, T. Fagni, and M. Tesconi, ‘‘CrisMap:
A big data crisis mapping system based on damage detection and geop-
arsing,’’ Inf. Syst. Frontiers, vol. 20, no. 5, pp. 993–1011, Oct. 2018, doi:
10.1007/s10796-018-9833-z.
[42] M. K. Hassan, A. I. El Desouky, S. M. Elghamrawy, and A. M. Sarhan,
‘‘Intelligent hybrid remote patient-monitoring model with cloud-based
framework for knowledge discovery,’’ Comput. Electr. Eng., vol. 70,
pp. 1034–1048, Aug. 2018, doi: 10.1016/j.compeleceng.2018.02.032.
[43] F. Gu, B. Ma, J. Guo, P. A. Summers, and P. Hall, ‘‘Internet of Things and
big data as potential solutions to the problems in waste electrical and elec-
tronic equipment management: An exploratory study,’’ Waste Manage.,
vol. 68, pp. 434–448, Oct. 2017, doi: 10.1016/j.wasman.2017.07.037.
[44] Y. Arfat, M. Aqib, R. Mehmood, A. Albeshri, I. Katib, N. Albogami, and
A. Alzahrani, ‘‘Enabling smarter societies through mobile big data fogs
and clouds,’’ Procedia Comput. Sci., vol. 109, pp. 1128–1133, Jan. 2017,
doi: 10.1016/j.procs.2017.05.439.
[45] S. Istephan and M.-R. Siadat, ‘‘Unstructured medical image query
using big data—An epilepsy case study,’’ J. Biomed. Informat., vol. 59,
pp. 218–226, Feb. 2016, doi: 10.1016/j.jbi.2015.12.005.
[46] D. Mourtzis, E. Vlachou, and N. Milas, ‘‘Industrial big data as a result
of IoT adoption in manufacturing,’’ Procedia CIRP, vol. 55, pp. 290–295,
Jan. 2016, doi: 10.1016/j.procir.2016.07.038.
[47] S. Rinaldi, A. Flammini, M. Pasetti, L. C. Tagliabue, A. C. Ciribini, and
S. Zanoni, ‘‘Metrological issues in the integration of heterogeneous lot
devices for energy efficiency in cognitive buildings,’’ in Proc. IEEE Int.
Instrum. Meas. Technol. Conf. (I2MTC), May 2018, pp. 1–6.
[48] P. Ta-Shma, A. Akbar, G. Gerson-Golan, G. Hadash, F. Carrez, and
K. Moessner, ‘‘An ingestion and analytics architecture for IoT applied to
smart city use cases,’’ IEEE Internet Things J., vol. 5, no. 2, pp. 765–774,
Apr. 2018, doi: 10.1109/JIOT.2017.2722378.
[49] A. C. Onal, O. Berat Sezer, M. Ozbayoglu, and E. Dogdu, ‘‘Weather data
analysis and sensor fault detection using an extended IoT framework with
semantics, big data, and machine learning,’’ in Proc. IEEE Int. Conf. Big
Data (Big Data), Dec. 2017, pp. 2037–2046.
[50] S. Balaji, M. Patil, and C. McGregor, ‘‘A cloud based big data based online
health analytics for rural NICUs and PICUs in india: Opportunities and
challenges,’’ in Proc. IEEE 30th Int. Symp. Comput.-Based Med. Syst.
(CBMS), Jun. 2017, pp. 385–390.
[51] L. Carnevale, A. Celesti, M. Fazio, P. Bramanti, and M. Villari, ‘‘How to
enable clinical workflows to integrate big healthcare data,’’ in Proc. IEEE
Symp. Comput. Commun. (ISCC), Jul. 2017, pp. 857–862.
[52] S. Amini, I. Gerostathopoulos, and C. Prehofer, ‘‘Big data analytics archi-
tecture for real-time traffic control,’’ in Proc. 5th IEEE Int. Conf. Models
Technol. Intell. Transp. Syst. (MT-ITS), Jun. 2017, pp. 710–715.
[53] Herrnansyah, Y. Ruldeviyani, and R. F. Aji, ‘‘Enhancing query perfor-
mance of library information systems using NoSQL DBMS: Case study
on library information systems of universitas indonesia,’’ in Proc. Int.
Workshop Big Data Inf. Secur. (IWBIS), Oct. 2016, pp. 41–46.
[54] I. Azimi, A. Anzanpour, A. M. Rahmani, P. Liljeberg, and T. Salakoski,
‘‘Medical warning system based on Internet of Things using fog com-
puting,’’ in Proc. Int. Workshop Big Data Inf. Secur. (IWBIS), Oct. 2016,
pp. 19–24.
[55] P.-Y. Wu, C.-W. Cheng, C. D. Kaddi, J. Venugopalan, R. Hoffman,
and M. D. Wang, ‘‘-Omic and electronic health record big data analytics
for precision medicine,’’ IEEE Trans. Bio-Med. Eng., vol. 64, no. 2,
pp. 263–273, Feb. 2017, doi: 10.1109/TBME.2016.2573285.
[56] E. Patti and A. Acquaviva, ‘‘IoT platform for smart cities: Requirements
and implementation case studies,’’ in Proc. IEEE 2nd Int. Forum Res.
Technol. Soc. Ind. Leveraging Better Tomorrow (RTSI), Sep. 2016, pp. 1–6.
[57] W. Seo, N. Kim, and S. Choi, ‘‘Bigdata framework for analyzing patents to
support strategic R&D planning,’’ in Auckland. IEEE, 2016, pp. 746–753.
[Online]. Available: https://ieeexplore.ieee.org/document/7588929
[58] V. Dineshreddy and G. R. Gangadharan, ‘‘Towards an, ‘Internet Things
framework for financial services sector,’’ in Proc. 3rd Int. Conf. Recent
Adv. Inf. Technol. (RAIT), Dhanbad, India, Mar. 2016, pp. 177–181.
[59] Y. Sun, H. Song, A. J. Jara, and R. Bie, ‘‘Internet of Things and big data
analytics for smart and connected communities,’’ IEEE Access, vol. 4,
pp. 766–773, 2016, doi: 10.1109/ACCESS.2016.2529723.
[60] S. Pirttikangas, E. Gilman, X. Su, T. Leppanen, A. Keskinarkaus,
M. Rautiainen, M. Pyykkonen, and J. Riekki, ‘‘Experiences with smart city
traffic pilot,’’ in Proc. IEEE Int. Conf. Big Data (Big Data), Dec. 2016,
pp. 1346–1352.
[61] A. Aguilera, R. Grunzke, U. Markwardt, D. Habich, D. Schollbach, and
J. Garcke, ‘‘Towards an industry data gateway: An integrated platform
for the analysis of wind turbine data,’’ in Proc. 7th Int. Workshop Sci.
Gateways, Jun. 2015, pp. 62–66.
[62] A. Abusharekh, S. A. Stewart, N. Hashemian, and S. S. R. Abidi,
‘‘H-DRIVE: A big health data analytics platform for evidence-informed
decision making,’’ in Proc. IEEE Int. Congr. Big Data, Jun. 2015,
pp. 416–423.
[63] M. Panahiazar, V. Taslimitehrani, A. Jadhav, and J. Pathak, ‘‘Empower-
ing personalized medicine with big data and semantic Web technology:
Promises, challenges, and use cases,’’ in Proc. IEEE Int. Conf. Big Data
(Big Data), Oct. 2014, pp. 790–795.
[64] A. Sebaa, F. Chikh, A. Nouicer, and A. Tari, ‘‘Medical big data ware-
house: Architecture and system design, a case study: Improving healthcare
resources distribution,’’ J. Med. Syst., vol. 42, no. 4, p. 59, Apr. 2018, doi:
10.1007/s10916-018-0894-9.
[65] J. F. Sánchez-Rada, A. Pascual, E. Conde, and C. A. Iglesias, ‘‘A big
linked data toolkit for social media analysis and visualization based
on W3C Web components,’’ in On Move to Meaningful Internet Sys-
tems. Valletta, Malta: Springer, 2018, pp. 498–515. [Online]. Available:
https://link.springer.com/chapter/10.1007/978-3-030-02671-4_30
[66] A. Smirnov, A. Ponomarev, N. Teslya, and N. Shilov, ‘‘Human-computer
cloud for smart cities: Tourist itinerary planning case study,’’ in Business
Information Systems Workshops (Lecture Notes in Business Information
Processing), vol. 303, W. Abramowicz, Ed. Cham, Switzerland: Springer,
2017, pp. 179–190.
[67] A. Aguilera, R. Grunzke, D. Habich, J. Luong, D. Schollbach,
U. Markwardt, and J. Garcke, ‘‘Advancing a gateway infrastructure for
wind turbine data analysis,’’ J. Grid Comput., vol. 14, no. 4, pp. 499–514,
Dec. 2016, doi: 10.1007/s10723-016-9376-9.
[68] R. S. Santos, T. A. Vaz, R. P. Santos, and J. M. P. de Oliveira, ‘‘Big
data analytics in a public general hospital,’’ in Machine Learning, Opti-
mization, and Big Data (Lecture Notes in Computer Science), vol. 10122,
P. M. Pardalos, P. Conca, G. Giuffrida, and G. Nicosia, Eds. Cham,
Switzerland: Springer, 2016, pp. 433–441.
186618 VOLUME 8, 2020
M. Volk et al.: Identifying Similarities of Big Data Projects—A Use Case Driven Approach
[69] H. Khazaei, S. Zareian, R. Veleda, and M. Litoiu, ‘‘Sipresk: A
big data analytic platform for smart transportation,’’ in Proc.
1st EAI International Summit, Smart City 360◦. Bratislava,
Slovakia: Springer, 2016, pp. 419–430. [Online]. Available:
https://link.springer.com/chapter/10.1007/978-3-319-33681-7_35
[70] A. Fiannaca, L. La Paglia, M. La Rosa, A. Messina, P. Storniolo, and
A. Urso, ‘‘Integrated DB for bioinformatics: A case study on analysis
of functional effect of MiRNA SNPs in cancer,’’ in Proc. Int. Conf. Inf.
Technol. Bio Med. Inform., Porto, Portugal, Sep. 2016, pp. 214–222.
[71] A. Majumdar and I. Bose, ‘‘Detection of financial rumors using big data
analytics: The case of the bombay stock exchange,’’ J. Organizational
Comput. Electron. Commerce, vol. 28, no. 2, pp. 79–97, Apr. 2018, doi: 10.
1080/10919392.2018.1444337.
[72] G. Escobedo, N. Jacome, and G. Arroyo-Figueroa, ‘‘Big data & analytics
to support the renewable energy integration of smart grids—Case study:
Power solar generation,’’ in Proc. 2nd Int. Conf. Internet Things, Big Data
Secur. IoTBDS, Porto, Portugal, Apr. 2017, pp. 267–275.
[73] Y. Zhuang, Y. Wang, J. Shao, L. Chen, W. Lu, J. Sun, B. Wei, and J. Wu,
‘‘D-ocean: An unstructured data management system for data ocean envi-
ronment,’’ Frontiers Comput. Sci., vol. 10, no. 2, pp. 353–369, Apr. 2016,
doi: 10.1007/s11704-015-5045-6.
[74] C.-M. Chen, ‘‘Use cases and challenges in telecom big data analytics,’’
APSIPA Trans. Signal Inf. Process., vol. 5, pp. 1–7, Dec. 2016, doi:
10.1017/ATSIP.2016.20.
[75] M. F. Huber, M. Voigt, and A.-C. N. Ngomo, ‘‘Big data architec-
ture for the semantic analysis of complex events in manufacturing,’’
in Informatik. Bonn, Germany: Gesellschaft fpr Informatik e.V., 2016,
pp. 353–360. [Online]. Available: https://dl.gi.de/handle/20.500.12116/
1139;jsessionid=D794018779FF36E5A6CBE13273EE9C67
[76] Q. Huang, G. Cervone, D. Jing, and C. Chang, ‘‘DisasterMapper,’’ in
Proc. 4th Int. ACM SIGSPATIAL Workshop Anal. Big Geospatial Data
BigSpatial, 2015, pp. 1–6.
[77] M. Xu, S. Siraj, and L. Qi, ‘‘A Hadoop-based data processing platform for
fresh Agro products traceability,’’ in Proc. Eur. Conf. Data Mining, 2015,
pp. 37–44. [Online]. Available: http://www.iadisportal.org/components/
com_booklibrary/ebooks/201508L005.pdf
[78] M. Steinbach, G. Karypis, and V. Kumar, ‘‘A comparison of document
clustering techniques,’’ in Proc. TextMining Workshop KDD, May 2000,
pp. 1–20.
[79] Y. Zhao, G. Karypis, and U. Fayyad, ‘‘Hierarchical clustering algorithms
for document datasets,’’ Data Mining Knowl. Discovery, vol. 10, no. 2,
pp. 141–168, Mar. 2005, doi: 10.1007/s10618-005-0361-3.
[80] L. Kaufman and P. J. Rousseeuw, Finding Groupsin Data: An Introduction
to Cluster Analysis, 99th ed. Hoboken, NJ, USA: Wiley, 2009. [Online].
Available: http://gbv.eblib.com/patron/FullRecord.aspx?p=469065
[81] M. Kantardzic, Data Mining: Concepts, Models, Methods, and Algorithms,
2nd ed. Piscataway, NJ, USA: IEEE Press, 2011.
[82] J. Cleve and U. Lämmel. Data Mining. München: De Gruyter Oldenbourg.
[Online]. Available: https://doi.org/10.1515/9783110456776
[83] P.-N. Tan, M. Steinbach, A. Karpatne, and V. Kumar, Introduction to Data
Mining. London, U.K.: Pearson, 2019.
[84] MATLAB. Agglomerative Hierarchical Cluster Tree—MATLAB
Linkage—MathWorks. Accessed: Mar. 27, 2020. [Online]. Available:
https://mathworks.com/help/stats/linkage.html?s_tid=mwa_osa_
a#d117e514451
[85] A. L. Marra, F. Martinelli, P. Mori, and A. Saracino, ‘‘Implementing usage
control in Internet of Things: A smart home use case,’’ in Proc. IEEE
Trustcom/BigDataSE/ICESS, Aug. 2017, pp. 1056–1063.
[86] G. Alfian, M. F. Ijaz, M. Syafrudin, M. A. Syaekhoni, N. L. Fitriyani, and
J. Rhee, ‘‘Customer behavior analysis using real-time data processing,’’
Asia Pacific J. Marketing Logistics, vol. 31, no. 1, pp. 265–290, Jan. 2019,
doi: 10.1108/APJML-03-2018-0088.
[87] K. Vassakis, E. Petrakis, I. Kopanakis, J. Makridis, and G. Mastorakis,
‘‘Location-based social network data for tourism destinations,’’ in Big
Data and Innovation in Tourism, Travel, and Hospitality: Managerial
Approaches, Techniques, and Applications, M. Sigala, R. Rahimi, and
M. Thelwall, Eds. Singapore: Springer, 2019, pp. 105–114.
[88] S. Muthuramalingam, A. Bharathi, S. Rakesh kumar, N. Gayathri,
R. Sathiyaraj, and B. Balamurugan, ‘‘IoT based intelligent transporta-
tion system (IoT-ITS) for global perspective: A case study,’’ in Internet
of Things and Big Data Analytics for Smart Generation, V. E. Balas,
V. K. Solanki, R. Kumar, and M. Khari, Eds. Cham, Switzerland: Springer,
2019, pp. 279–300.
[89] J. vom Brocke and A. Maedche, ‘‘The DSR grid: Six core dimensions for
effectively planning and communicating design science research projects,’’
Electron. Markets, vol. 29, no. 3, pp. 379–385, Sep. 2019, doi: 10.1007/
s12525-019-00358-7.
[90] M. J. Zaki, Eds., On Finding the Natural Number of Topics With Latent
Dirichlet Allocation: Some Observations: Advances in Knowledge Discov-
ery and Data Mining. Berlin, Germany: Springer, 2010.
MATTHIAS VOLK (Graduate Student Member, IEEE) studied business
informatics at the Faculty of Computer Science, Otto von Guericke Univer-
sity Magdeburg (OVGU). He received the master’s degree in 2016. He is
currently pursuing the Ph.D. degree. Since then, he has been employed
as a Scientific Researcher. During his studies, he gained lots of practical
experience as a software developer in different companies such as Volk-
swagen. During his scientific career, he participated in many international
scientific congresses and projects, not only as a speaker but also as a reviewer
or a session chair. His research interests include domain of data-intensive
systems, related projects, technologies, and the management of them.
DANIEL STAEGEMANN studied computer science at Technical University
Berlin (TUB). He received the master’s degree in 2017. He is currently
pursuing the Ph.D. degree with the Otto von Guericke UniversityMagdeburg.
Since 2018, he has been employed as a Scientific Researcher with OVGU.
His research interest includes big data, especially the testing.
IVAYLA TRIFONOVA studied business informatics at the Otto von Guericke
University Magdeburg. She received the master’s degree in 2019. She is
currently working as an IT Consultant in the area of Life Science at a large
European consulting and IT services company.
SASCHA BOSSE studied computer science at the Faculty of Computer
Science, Otto von Guericke University Magdeburg. He received the mas-
ter’s degree in 2011, and the academic degree Doktoringenieur in 2016.
Since 2012, he has been working as a Researcher with the VLBA Lab.
Since 2020, he has also been working as a Subject Specialist for computer
science and mathematics with University Library Magdeburg, where he is
also responsible for the business applications. His research interests include
IT service management, modeling, simulation, and optimization.
KLAUS TUROWSKI studied business and engineering at the University of
Karlsruhe. He received the Ph.D. degree from the Institute for Business
Informatics, University of Münster, and the habilitated degree in business
informatics from the Faculty of Computer Science, Otto von Guericke Uni-
versity Magdeburg. In 2000, he deputized the Chair of business informatics
at the University of the Federal Armed Forces München. Since 2001, he has
been heading the Chair of business informatics and systems engineering with
the University of Augsburg. Since 2011, he has also been heading the Chair
of business informatics (AG WI) with the Otto von Guericke University
Magdeburg, the Very Large Business Applications Lab (VLBA Lab), and the
world’s largest SAP University Competence Center (SAP UCC Magdeburg).
Additionally, he worked as a guest lecturer at several universities around the
world. He was a Lecturer with the Universities of Darmstadt and Konstanz.
He was a (co-)organizer of a multiplicity of national and international
scientific congresses and workshops (>30) and acted as a member of several
programme commitees (>130), and expert Groups. In the context of his
university activities as well as an independent consultant he gained practical
experience in industry.
VOLUME 8, 2020 186619
