Content uploaded by Rashi Jaiswal
Author content
All content in this area was uploaded by Rashi Jaiswal on Dec 04, 2023
Content may be subject to copyright.
-187-
A REVIEW ON TEMPORAL DATA MINING
Rashi Jaiswal*, Brijendra Singh**
Department of Computer Science, University of Lucknow, Lucknow (U.P.), India.
E-mail: rashijaiswal.rj95@gmail.com*, drbri_singh@ hotmail.com**
ABSTRACT
Temporal data mining is much more complex because additional challenges such as data handling, modelling,
etc. exist during the computation process of temporal data. This paper categorizes Temporal Data Mining into
three leveled structuresas data, model, and system. We explore the temporal data properties at data level and
various traditional and advanced models at the model level. We examine them by experiments to analyze the
appropriate method for different problem handling. It also discusses the real-world applications of temporal data
mining at the system level in different domains. This review also provides the vision of the future research
direction related to data, model, and system-level by shedding some light on challenges in the complete data
mining process.
Keywords : Data Mining Techniques, Temporal Data, Learning Models, Data, Model, System, Information
Retrivel.
1. INTRODUCTION
In the real world, most of the data is generated in temporal form. It is very important to compute the
data and retrieve the information. Temporal data are sequences of a primary data type relating to time
instances, most commonly categorical or numerical values, and sometimes composite or multivariate
information. The computation of quality-oriented data provides a more accurate result or more accurate
information. Information can be retrieved or predicted from temporal data by using data mining
techniques[1][2]. In literature, various researchers have given generic survey articles and an overview of
data mining methods [3][4][5][6][7][8]. Other review articles focus only on one type of method based on
application, technical approach, or data type[9]. Data mining methods can be classified according to the
domain-based application. Most of the survey articles are only focus on single data mining
techniquessuch as subjective association rule mining for temporal sequences [10], Data Clustering[11],
Anomaly detection[12], Clustering algorithms[13], Outlier detection for temporal data[14], association
rule mining[15]. This paper presents a review of data mining techniques for temporal datasets.This paper
has reviewed some popular methods for data mining and specially focused on tem-poral datasets
computation. In the current world the data having the main factor which is 'Time' which affects the
information retrieval process and purposes. It is the biggest challenge in today's world for performing
real-world data and does an accurate prediction for the decision-making process. This paper categorized
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-188-
the temporal data mining process based on three-level of the sub process. Through Data Mining, extract
useful information in given datasets by extract patterns and identify relationships. Various data mining
problems arise at the different level of mining process, problems at data level, model level and system
level. After reviewing the various research papers on temporal data mining and by using the different data
mining techniques, solving the various temporal data mining problems arise at different levels of the
mining process. In this paper, the problem during the process described as three levels have been given
for temporal data mining:
1. At the Data level, data generation and acquisition task-based problems arise. Here the data
preprocessing and work on the temporal data quality maintenance and assurance, based on the temporal
data features where the data is going through the process for the clean the data and impute the missing
value, and also balancing the class imbalanced datasets[2]. The problem can be solved through the Data
and feature engineering [23] in the data mining process.
2. Problem at the Model level is to manage and classify or the computation process for the tem-poral
data mining has been done through the different models. The performance of the models has been
calculated based on performance metrics and the execution time taken for the tem-poral data mining. The
computation impact to the model performance.
3. At the System level, is the information is relevant or not? It's all depends on system perfor-mance
which is an integration of data and model work performance, when the learning process on various types
of data for generating relevant information through needful automated system methodology as well as a
traditionally manual process[16] for the various application-based system in various domain.
This research paper focused on temporal data mining and included temporal data mining methods
comparison with experimental analysis.
This paper is organized as follows: Section 1 introduces the contribution of the paper that is temporal data
mining categorized into three levels Data, Model, and System level. In Section 2, identification of
temporal datasets with required parameters and focused on the process of identification of the required
parameters for solving the particular problem in the temporal data-sets. Section 3 provides the temporal
data mining at model level in detail and also provides an experimental analysis of temporal data mining
algorithms with the comparison of models concerning the execution time required for the temporal data
mining. Section 4 discussed the applications of temporal data mining that presents about the system level.
Section 5 discussed the proposed work with findings of this study on temporal data mining The data is
temporal for the mining then various issues and challenges occur in computation. The details have been
discussed in the results as findings of this study. And Section 6 concludes the paper with the future scope.
Important nota-tions/abbreviations have given in Table 1.
A Review on Temporal Data Mining
-189-
Table 1.Notations/Abbreviations
Notation/Abbreviations Description/Full name
RNN Recurrent neural network
CNN Convolutional Neural Network
LSTM Long Short Term Memory Network
BiLSTM Bidirectional Long Short Term Memory Network
GRU Gated Recurrent Network
DT Decision Tree
KNN K Nearest Neighbor
NN Neural Network
BPTT Back propagation through Time
FCM Fuzzy C means
BIRCH Balanced Iterative Reducing and Clustering using Hierarchies
CLARA Clustering Large Applications
GCM General c-means clustering
LASSO Least Absolute Selection Shrinkage Operator
MSE Mean Square Error
R2 R-Square
2. TEMPORAL DATA MINING AT DATA LEVEL
Time is the important attribute in the temporal data mining. Data changes with time which reflect the data
analysis and data modeling techniques,[17][18][19]. Whenever, Temporal aspects in the mining part, have
to take the historical data for the prediction or the time-oriented applications then temporal data modeling
is needed to handle the temporal datasets with some pattern and sequences extraction functionality
[20][21][22][23][24]. So, for handling the temporal data in data mining there are various techniques
available that can help to handle the temporal data mining in an efficient manner [25]. The temporal
datasets have a minimum of a single feature that is related to temporal properties or changes according to
temporal variation. Some temporal data properties make any dataset time effective. Four broad categories
of temporality within data can be deter-mined by Roddick et al. in [26] are Static, Sequences, Time-
stamped, and Fully-temporal. Theo-phano-Mitsa [27] has described in her book with deep literature on
temporal data mining. There are three types of data: (i) Time series, (ii) temporal sequences, and (iii)
Semantic temporal data. An event can be considered as a special case of a temporal sequence with one
time-stamped element.
The rapid development of technology has led to the registration of many processes in an electronic
environment, the storage of these records, and the accessibility of these records when re-quested. The
accumulation of a large amount of data stored in databases and the process of parsing and screening
useful information made data mining necessary. The data preprocessing and the data cleaning process has
done at data level by applying the different sub-process to gain the quality based input for the mining
process. So, there are some different methods available for dealing with temporal data because temporal
data having different properties than normal data [28][29]. The real world datasets, having various
challenges related to data handling. The temporal data analysis phase in the temporal data mining process
has different issues which are handled through the data preprocessing and imputation process to make the
data clean and relevant [30][31][32].
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-190-
Table 2. Some Temporal Datasets with features details
S. No. Datasets Data type No. of Features Temporal Properties
1 Shampoo Sales Dataset [33] Univariate 36 observations 2 attributes Day time
2 Minimum Daily Temperatures Dataset [34] Univariate 3650 observations 2 attributes Date
3 Monthly Sunspot Dataset [35] Univariate 2820 observations 2 attributes Months
4 Daily Female Births Dataset [36] Univariate 365 observations 2 attributes Date time
5 EEG Eye State Dataset [37] Multivariate 14,980 observations 15 attributes Event
6 Occupancy Detection Dataset [38] Multivariate 20,560 observations 715 attributes Date , Time
7 Ozone Level Detection Dataset [39] Multivariate 2,536 observations 73 attributes Time stamp
8 Wine quality [40] Multivariate 4898 observations 12 attributes Timestamp
9 Credit card default [41] Multivariate 30000 observations 24 attributes Event based
10 London merged [42] Multivariate 13550 observations 7 attributes Timestamp
11 Gracious datasets [43] Multivariate 38766 observations 3 attributes Date and time
12 Sales transaction weekly dataset [44] Multivariate 812 observations 82 attributes Timestamp (in weeks)
13 Bread Basket DMS Dataset [45] Multivariate 21294 observations 4 attributes Date
14 Online Retail Dataset [46] Multivariate 541910 observations 8 attributes Date and time
15 Boston crime Dataset [47] Multivariate 501071 observations 19 attributes Timestamp
16 House power consumption dataset [48] Multivariate 1048576 observations 9 attributes Timestamp
17 Airline's passenger Datasets [49] Univariate 145 observations 2 attributes Monthly Timestamp
18 Air Quality UCI Dataset [50] Multivariate 9358 observations 10 attributes Date
19 Weather AUS Dataset [51] Multivariate 142194 observations 24 attributes Date
20 Bike-sharing Dataset [52] Multivariate 17380 observations 14 attributes Time and Date
21 KC House Dataset [53] Multivariate 21614 observations 21 attributes Date (Day-wise)
22 Gold Price Dataset [54] Univariate 2290 observations 7 attributes Date (Day-wise)
Table 1.Relationship between Temporal Data Mining Techniques and Types
S.no. Data mining Techniques Temporal data types
Event sequences Regular time series Timestamps Fully Temporal
1 Association Rule [10]
2 Classification [55]
3 Clustering [15]
4 Outlier Detection [56][14]
5 Regression [57]
The temporal datasets have categorized into two types according to the temporal aspects dependencies:
Univariate and Multivariate [58][59]. According to that the data modeling and further operation on
datasets can be done for increasing the performance of the system. The aspects of temporal data are
usually included valid time, transaction time, or decision time. The temporal data is related to
time/periods with no ends which always maintained the transaction time with temporal primary keys or
time attributes. The temporal data representation is always having the approach where the time is a basic
entity and the other attributes are dependent on them. The temporal data can be categorized with the given
properties which relate or deals with the time-based enabled entity such as: Date-time[60], Timestamp
data[61][62][63], Interval Based data [64][65][66], Time-series Data [67][68][69] and Event-based
data[70][71][72].
For selecting the correct features for the temporal based model to compute the quality based input and
A Review on Temporal Data Mining
-191-
getting more accurate results after applying the data mining techniques on these quality input data [55]. It
is essential to select the temporal features by characterization of the techniques for the further imputation
process. The temporal features extraction can be done from the temporal data through Date related
features, Time related features, Lag Features, Rolling window, and Expanding window. The data
preprocessing [73] task perform on the temporal data based on their feature sets and properties.
Preprocessed data having quality which helps to increase the performance of the temporal data mining
methods for better results.
The more refined data gives more accurate results and makes the model perform better and give a more
accurate result. The temporal data having different properties and characteristics which makes the
computation process for the mining is different. Temporal datasets should have a single dependent or
independent time-based feature.
The datasets may be synthetic (Artificial generated datasets), augmented (Manipulated data-sets), and also
Real (online data/ real-world data) datasets. There are many open sources for fetching the datasets for
experimental study or research purposes [28][74]. The datasets had taken from the various machine
learning repository (UCI Machine Learning Repository, Kaggle).This paper has taken the temporal
datasets from the Kaggle and UCI ML repository. The list of some temporal benchmark datasets has
given in Table 2 with their properties in detail. This is very important to analyze the data before use. The
temporal data totally depends on the temporal attribute so, it's hard to do preprocessing and manipulate in
it. The fluctuation in timestamps can be do for manipulate or generate the new patterns from same
datasets by resampling techniques. Where decomposition of data based on its components can be do as
per need to validate the outcomes. This survey is an attempt to solve the problem by providing a structure
for data mining process where the problems at the data level, found that the temporal features have
prioritized for the temporal data mining so, it should be clarified that at least single attribute having
temporal property.
The temporal features are essential for temporal information retrieval process and other features in the
dataset may be dependent or independent on temporal attribute. The feature engineering has to apply for
the necessary feature extraction in the temporal datasets and also dimensionally reduction process applied
on the large dataset for fast computation in further mining process. Before using the datasets for data
mining process should be sure that the data is complete and clearly sophisticated or not. It is easy to work
on uni-variate data in temporal data mining. However, the computing complexities increase with the
multivariate datasets and have to use some different algorithms to handle them. Such as Principal
component analysis used for prepare data with important features and normalization of data as in combo.
But for temporal data, PCA and missing value imputation is not much suitable due to unbreakable
patterns, and dependent temporal sequences of temporal data. This is another issue with temporal data.
So, before computation, we have to resolve data level issues. In future research work, the problems related
at the data level related to temporal attributes in data mining process can be solved through developing
technique that works on the imperfect real data for getting the quality data.
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-192-
3. TEMPORAL DATA MINING AT MODEL LEVEL
At this level, the dataset has been taken as input for the model processing to information retrieval get
as an output [16]. The clean and refined input of temporal data has increases the performance of the
models. The method for each mining task has selected which is based on previous section of this paper.
The specific methods have existed for doing the temporal data mining tasks with the different
characteristics and properties of temporal data for the computation process[75][76].
3.1 Temporal Models for Data Mining
Temporal data computation has been done through the temporal data models [27] which are work-ing
with the mining-based methodology and the methods. The supportive data mining techniques are applying
to the temporal data for the computation process. Based on temporal data properties, the computational
model has been chosen for the imputation of input data for retrieving the in-formation. The temporal
models can be characterized by the data mining techniques and the types of temporal data which has
given in Table 3 through the relationship between them.
Various statistical [77] and machine learning [24][76][78] and the deep learning approach [79] based
methods have existed for the temporal data mining process. This paper has selected some data mining
methods for comparing and analyzing them through their performance measurement based on different
performance metrics and execution time of the specific model. For each data mining task, we have
selected their specific methods for the mining process on the temporal datasets. This paper have done all
experiment with python code and used the python based machine learning library: SK-learn (used for
Data mining models), and many other libraries like pandas and NumPy for the computation of data and
matplotlib and seaborn for the visualization of the data and the output.
3.2 Performance evaluation of Data Mining Methods
Performance Evaluation of all selected methods has been done through their performance metrics-based
evaluation and also measures the time taken in the execution process. Performance metrics are used as
data representation and the overall quality of the computation methods and methodology. In this section, a
comparison of all methods based on performance metrics for quality checks has been presented.
I. Performance evaluation for the selected Association Rule mining
The Association rule mining process can be done through different methods. Various researchers have
given various methods for temporal association rule mining [80][81][82][83][84] with the ability to
generate time-dependent patterns between the huge amounts of temporal data [85]. As per based on the
popularity of the FP growth and A priori Algorithm, we have selected both algorithms for the experiments
of association rule mining on temporal datasets. Temporal association rule mining has been done for the
pattern recognition in the temporal datasets to measure the performance of selected models as per the
measurement of the support, confidence, and the lift score card. In this paper, we find the execution time
used to compare and analyze the algorithms.
After going through the experiments based on performance evaluation of the selected Association Rule
mining models on two datasets (Groceries and Bread Basket Dataset), found that FP growth performs
A Review on Temporal Data Mining
-193-
well than Apriori Algorithm. In our experiment, the top one association rule for each dataset has been
taken for comparison, and the result is shown in table 4 and visualize the score in Fig. 1.
Table 4.Performance evaluation of Association Rule Method
Datasets Methods Support Confidence lift Time of Execution
(seconds)
Groceriesdataset FP Growth 0.0065 0.1253 0.7936 0.2067
Apriori 0.0082 0.1213 0.7681 1.5311
Bread Basket Dataset FP Growth 0.0894 0.2751 0.5792 0.1002
Apriori 0.0065 0.2255 0.6939 0.4980
Fig.1. Performance of Association Rule Models
II. Performance evaluation for the selected Classification methods
Temporal data classification is an evolving area in machine learning and data mining in which, time is
included in the learning procedure. The temporal data mining Classification techniques has selected based
on on-demand, the selected algorithms are Gaussian Naïve Bayes, Support Vector Machine, Decision
Tree, Random Forest, KNN, MLP, LSTM, RNN, and GRU for the experimental study of the
classification of temporal datasets has done. Where the performance metrics decided the model's
capability for the main metrics of the classification model are accuracy, precision, recall, f1_score, and
support. The novel parameter has also been calculated to measure the performance of the models which is
the execution time, with the computation of the other parameters used for calculating the performance
score for better results and selecting the best algorithm among them. The performance metrics score with
the execution time of the classification models for the input temporal datasets has given in Table 5. Where
it has cleared that model with max accuracy and min execution time is more effective and useful than
others.
The experimental study on the selected classification model has done on the three temporal datasets
(London merged dataset, Monthly car sale Dataset, and Daily Min. Temperature Dataset) with the
different data properties. The experimental results found that the SVM has taken more execution time as a
classifier whereas KNN as a traditional model performs better with less execution time but the accuracy
of both models has approximately the same, as shown in Table 5. The deep learning models have given
the best result incomparison to other selected methods. The deep learning models like RNN, LSTM, and
GRU take much time to compute due to complex neural network architecture but they facilitate to
hypertuned the models as per need. So, it is easy to compute and fit the model with a deep learning
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-194-
approach. Deep learning models give more accurate results with a large number of datasets in less time in
comparison to traditional models.
Table 5. Performance Evaluation of Classification Methods
Datasets Methods Precision Recall F1_score Accuracy Time of
Execution
London
merged
Dataset
Gaussian Naïve Bayes 0.8363 0.8219 0.8243 0.8219 0.1100
Support Vector Machine 0.8776 0.8765 0.8769 0.8765 299.9200
Decision Tree 0.8120 0.8013 0.8036 0.8013 2.2200
Random Forest 0.8933 0.8794 0.8810 0.8794 0.5800
KNN 0.8715 0.8524 0.8545 0.8524 0.4000
MLP 0.8776 0.8765 0.8769 0.8765 20.1200
RNN 0.9026 0.8863 0.8878 0.8863 140.8835
LSTM 0.8909 0.8857 0.8867 0.8857 278.3400
GRU 0.8733 0.8713 0.8720 0.8713 309.8420
Monthly car
sales Dataset
Gaussian Naïve Bayes 0.8175 0.8125 0.8118 0.8125 0.0010
Support Vector Machine 0.8750 0.8750 0.8750 0.8750 0.0020
Decision Tree 0.7915 0.7812 0.7793 0.7813 0.0015
Random Forest 0.8451 0.8438 0.8436 0.8438 0.0200
KNN 0.8810 0.8750 0.8745 0.8750 0.0010
MLP 0.8451 0.8438 0.8436 0.8438 0.1995
RNN 0.8750 0.8750 0.8750 0.8750 13.0556
LSTM 0.8810 0.8438 0.8398 0.8438 16.8182
GRU 0.8175 0.8125 0.8118 0.8125 21.4617
Daily Min.
Temperature
Dataset
Gaussian Naïve Bayes 0.8367 0.8365 0.8365 0.8365 0.0050
Support Vector Machine 0.8571 0.8566 0.8566 0.8566 0.4718
Decision Tree 0.7654 0.7653 0.7652 0.7653 0.0822
Random Forest 0.8441 0.8438 0.8438 0.8438 0.0882
KNN 0.8257 0.8256 0.8256 0.8256 0.0070
MLP 0.8277 0.8274 0.8274 0.8274 4.9281
RNN 0.8575 0.8575 0.8575 0.8575 26.8013
LSTM 0.8585 0.8584 0.8585 0.8584 100.8901
GRU 0.8438 0.8438 0.8438 0.8438 108.9737
Fig.2. Performance of classification Models
III. Performance evaluation for the selected Clustering methods
Clustering techniques are used for data engineering for data mining purposes[56][86][87]. The tem-poral
A Review on Temporal Data Mining
-195-
data clustering has been done through the basic method which is the hidden Markov model by w. Lin for
similar and periodic pattern finding in discrete time series data [88]. The K-means, f-cmeans, and BIRCH,
methods have been selected for the experimental-based comparison analysis for temporal data mining.
The performance score analysis has done based on Homogeneity, completeness, v-measure, and number
of clusters for each method. The comparison has been done for the execution time that has been taken for
the method's computation process.
Table 6.Performance Evaluation of Clustering Methods
Datasets Methods Number of
Clusters
Homogeneity Completeness V-measure Time of
Execution(seconds)
London
merged
Dataset
K-means 3 0.8484 0.7907 0.8186 0.2176
f-cmeans 3 0.7589 0.6910 0.7234 0.5063
BIRCH 3 0.5611 0.4631 0.5074 23.0921
K-means 5 0.6805 0.7012 0.6907 0.3850
f-cmeans 5 0.6984 0.7161 0.7071 1.1545
BIRCH 5 0.6696 0.7205 0.6941 21.5350
Daily min
temperatures
Dataset
K-means 3 0.5163 0.4759 0.4953 0.0797
f-cmeans 3 0.5163 0.4759 0.4953 0.0637
BIRCH 3 0.4219 0.5852 0.4903 0.0852
K-means 5 0.6639 0.6938 0.6785 0.1389
f-cmeans 5 0.6913 0.7231 0.7069 0.2106
BIRCH 5 0.4132 0.6225 0.4967 0.0812
monthly-car-
sales
Dataset
K-means 3 0.4996 0.4824 0.4909 0.0476
f-cmeans 3 0.5663 0.5496 0.5578 0.0070
BIRCH 3 1.0000 1.0000 1.0000 0.0130
K-means 5 0.9344 0.9381 0.9362 0.0441
f-cmeans 5 0.9343 0.9381 0.9362 0.0200
BIRCH 5 1.0000 1.0000 1.0000 0.0140
Fig.3. Performance of Clustering Model
As shown the experimental results of performance evaluation of selected clustering models on the
temporal datasets in table 6 also visualized in fig. 3, where we found that the number of clusters impacts
on model performance accordingly. In which, BIRCH has taken much more execution time than the K-
means. So, for dealing with large datasets K-means performs well with a high V-Score within less time.
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-196-
IV. Performance evaluation for the selected Outlier Detection
Outlier detection is used for anomaly detection in the datasets which fluctuates the efficiency of the
computational process of data mining[14]. The popular methods for the outliers detection process in temporal
data mining selected for experimental study are Isolation forest, One-class SVM, and Elliptic Envelope
detection. The performance metrics such as precision, recall, f1-score, Accuracy, and execution time have
been taken for the comparative analysis for deciding the performance scale for them.
The performance of selected methods has been analyzed through the experimental study by the comparing
the evaluated performance score of selected Outlier detection models with the Local factor Outlier
Detection as a base model for getting the performance score of Isolation Forest, One-class SVM, and
Elliptic Envelop model’s score shown in table 7. Where found that EE has taken less execution time and
performs well than the other selected models with the large datasets but the Isolation Forest has also given
more accurate results but taken more execution time than the EE. The visualization of performance Scores
has shown in fig.4.
Table 2. Performance Evaluation of Outlier Detection Methods
Datasets Methods Precision Recall F1-Score Accuracy Time of Execution
(seconds)
London
merged
Dataset
Isolation Forest 0.7772 0.7777 0.7774 0.7777 1.4659
One class SVM 0.7758 0.5379 0.6133 0.5379 24.6926
Elliptic Envelope 0.7765 0.7765 0.7765 0.7765 0.0140
Daily min
temperatures
Dataset
Isolation Forest 0.7580 0.7586 0.7583 0.7586 0.5663
One class SVM 0.7881 0.7625 0.7746 0.7625 0.2536
Elliptic Envelope 0.7578 0.7578 0.7578 0.7578 0.0080
monthly-car-
sales
Dataset
Isolation Forest 0.9444 0.9444 0.9444 0.9444 0.2015
One class SVM 0.8333 0.4815 0.5537 0.4815 0.0030
Elliptic Envelope 0.9074 0.9074 0.9074 0.9074 0.0300
Fig.4. Performance of Outlier Detection Models
V. Performance evaluation for the selected Regression methods
Regression is the technique for doing predictions based on past data [89]. In the temporal data mining, the
prediction process for future events has been done through the regression techniques appliedto the
temporal datasets as historical data and processed by the selected methods based on the related work.
A Review on Temporal Data Mining
-197-
When temporal data has been taken as an input for the future prediction then each model has performed
differently which has been measured through the performance metrics evaluation. The mean square error
and the R2_score have been computed for all used methods in the experiments. The details have given in
table 8. The method's performance has also been mentioned and measured concerning the execution time
of each method for comparison and analyzing the better once in between them.
The experimental study of regression models on the temporal datasets found that the deep learning models
(RNN, LSTM, GRU) have outperform and take much time with high R square scores. Bayesian and
Ridge have taken approximately the same as per R2 Score measurement but also found that if the amount
of data is large then the ridge performs well. Otherwise, deep learning models give good results with a
large amount of data.
Table 8. Performance Evaluation of Regression Methods
Datasets Methods MSE R2_Score Time of Execution
(seconds)
London merged
Dataset
Linear Regression 0.4618 0.6891 0.4231
Lasso Regression 0.4660 0.6833 5.0249
Ridge Regression 0.4617 0.6892 0.0913
Bayesian Regression 0.4616 0.6894 0.9571
LSTM 0.4020 0.7644 172.3227
RNN 0.3948 0.7728 103.8063
GRU 0.4471 0.7084 183.5200
Monthly car
sales Dataset
Linear Regression 0.3240 0.7063 0.0030
Lasso Regression 0.3204 0.7127 0.0055
Ridge Regression 0.3187 0.7156 0.0030
Bayesian Ridge 0.3140 0.7241 0.0110
LSTM 0.3545 0.6482 8.0563
RNN 0.3875 0.5798 12.7681
GRU 0.3512 0.6548 14.0341
Daily Min.
Temperature
Dataset
Linear Regression 0.3881 0.6713 0.0055
Lasso Regression 0.3878 0.6720 0.0890
Ridge Regression 0.3881 0.6713 0.0050
Bayesian Ridge 0.3882 0.6712 0.0070
LSTM 0.3872 0.6728 37.0174
RNN 0.3932 0.6625 17.6017
GRU 0.3865 0.6741 34.3644
Fig.5. Performance of Regression Models
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-198-
In this section, the experimental study has been done by applying the data mining techniques on temporal
datasets and evaluating the performance of the selected methods for analyzing the temporal data mining
methods’ performance. The results are shown with the performance score findings of selected methods for
temporal data mining in table 4, 5, 6, 7, and 8 respectively. According to the model’s performance score
evaluation, we have visualized the performance analysis-based charts fig. 1,2,3,4, and 5 for each data
mining technique respectively.
The experimental study for the temporal data mining has done by working on the temporal datasets taken
as an input for the implementation done with the some selected temporal models based on their popularity
for the analysis of different data mining techniques for the temporal data mining process and get the
output concerning the temporal execution of the models. The comparison analysis of existing temporal
information retrieval methods which is based on the performance evaluation has been done through the
experimental study of temporal data mining.
Overall, the model level of temporal data mining process where we select some traditional and advanced
approach based models to compare and analyze them and find FP growth for association rule mining,
SVM, KNN as traditional models and RNN, LSTM, and GRU as advanced models for classification,
BIRCH for clustering, Elliptic Envelop and Isolation forest for outlier Detection and Bayesian and Ridge
as traditional and recurrent neural network models for Regression methods outperform on temporal data
concerning time and accuracy which take less execution time and gives the better results respectively for
temporal data mining which has been illustrated. The statistical, machine learning and deep learning
approach-based methods have been selected for the relevant experimental study. Experimental Study
shows that the statistical methods perform well on small datasets otherwise computation of process
become complex and machine learning approach based methods perform better on large datasets than the
statistical methods but when we have multi-dimensional datasets then, the deep learning approach
performs better whereas GRU is faster on large scale datasets but the LSTM gives more accurate results.
During the experimental study, this survey found a big problem at the model level that the appropriate
method selection is the main task for any computation process in the data mining process. The other issue
at this level found that sometimes existing methods are not able to handle the task which included two and
more types of problems that impact to results, then need some new technique or mingled features of two
to four methods to solve a particular problem then ensemble methods[90] can handle this problem. The
contradiction existed between the methods' applicability because sometimes the computational time
matters and other side reliability also matters. So, it requires the attention of researchers as per the
requirement and the demand of methods in industries or the real world for the problem solving or method
selection at the model level in the data mining process.
4. TEMPORAL DATA MINING AT SYSTEM LEVEL
The various temporal data-based applications work on the system level implementation for temporal data
mining [91]. In which, the integration of the data and the model helps to develop a system for temporal
A Review on Temporal Data Mining
-199-
data mining and retrieve the information regarding the same [92]. The information retrieval process can
be done through the data visualization process and as well as numerical computation. The temporal
information can be viewed with the graphical representation as a chart and the graphs as an application at
the user end. We can also show it with the table format and also with the text values. The various
applications of temporal Data mining are available at the system level in different domains.
4.1 Prediction or forecasting
Forecasting is a technique that uses historical data as input to predict future trends. The prediction or the
forecast is done through temporal data mining for getting information regarding future event behavior.
There are so many applications that are available for the main purpose of predicting temporal data. For
Example, weather forecasting [93] and sale prediction [94].
Temporal datasets are used for weather forecasting (i.e. next three-day temperature prediction and air
quality of current location). Such types of applications are helpful for the planning of day-to-day life and
also alert for upcoming disasters (like- heavy rain, flood, hot or cold weather, thunderstorms, humidity,
and much other information about the weather). Here, the problem exists in the data collection and feature
extraction phase in the forecasting process. So, there is a need to provide fresh quality data of a particular
environment as well as historic data of whether a condition of a specific place and need to work with the
collaboration of spatial-temporal data. In agriculture, a weather forecasting system helps to get the
information regarding farming trends with the suitable situation and climate. It can be more helpful and
productive with the collaboration of remote sensing satellites based data for more effective and instant
results.
Similarly, sales forecast also worksthe same as the weather forecast strategy but it depends on the
organization's requirements as well as the property of product and consumer behavior. The temporal data
take as an input in the sales forecast system because the input data behavior change as per the trends and
seasonality dependency. The input data for the sales forecast system can get from the Customer
Relationship Management System of the particular organization and the results predict accordingly. Here,
find that the problem, a single system doesn’t fit into all domains. So, need to make a generalized system
which flexible as per the organization's environment and requirements with high performance.
The need for temporal data as input for the prediction process of future sales or in whether forecast
Systems. The System for temporal data computation of prediction process via computer with the spice of
artificial intelligence has helped to give more accurate results for the different tasks in domains.
4.2 Anomalies detection
Anomaly Detection is the activity ofthe identification of unexpected events or objects that are
irregularities or oddities. Sometimes datasets contain some odd or suspicious data which decreases the
accuracy rate of the output with difficulties in the initial computation process of data mining [12][13]. The
real-world data contains anomalies that impactthe mining process at the system level. So, it is essential to
analyze the data based on outliers where work on data by finding out the outliers and removing or refining
them. Anomaly detection is used for fraud detection, detecting the odds in something or some events (i.e.
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-200-
human behavior in a public place, system, plants and animals behavior characterization), etc. Today, so
many frauds are increasing like fake calls or account hacking, or unauthorized money transactions. Some
techniques [95] are available for fraud detection based on temporal data of transaction details. Such other
Systems, suspicious activity detection has done through the anomalies detection process from the
multimedia (live or recorded) data ina public place. Here, the problem exists that the crime is noticed after
it has been executed by a suspicious person. So, this issue can be solved by doing the work on the live
stream data with object detection and alert mode extensions. The temporal data-based Internet of Things
systems should involve making that kind of System from which to get instant results and directly impact
human security and privacy.
4.3 Trend analysis
In the current world, it is very necessary to the analysis of the trending pattern of the market to continue
the growth of the business which is helpful to launching a new product in the market for better results and
having more profit from the same product and increase the sale of the product [69]. Trends define the
market's current demands in the different domains through pattern analysis[96]. It can analyze the trends
through social networking connections and also on the bases of searching history of the consumer where
temporal data plays a very important role for analyze the customer behavior and interest and showing
them the ads of related other products. In this way, these applications are very helpful for market demand
analysis and increase sales also [21]. In the current world, the online system makes human life easy forthe
user side and also for the online business owner. The organization makes more profit on online systems
by applying artificial intelligence to that system which helps them to find the market trend. Some
problems in trend analysis are also existing that are exact pattern finding and the sudden pattern change as
per the demand in public. So, there needs to more research work on the current situation-based time series
data with the collaboration of event finding techniques with pattern analysis.
4.4 Report generation
The past and present data help to predict a future event. After analyzing the temporal data, it can be mined
and used for the report generation which reflects as information through an application. Such as some
application only reports the current situation of the time-based events based on previous historical data,
whereas some application has past and add current data for target report generation in any organization,
some such applications are used for student performance and behavior report generation based on their
time to time activities in the class [97][98]. At present time, online classes are the trend that has been
adopted by many schools and colleges. For that, the various web-based system available in which report
card generation module was added and performance analysis part also implemented. But here need to
make the whole education system virtual with the automated facility. So, the researcher should focus on
that kind of system formation process with the collaboration of data mining and artificial intelligence
where the machine should also have the capability to generate questions and answer as per the student
performance and fluctuate the study hour as per the need with time bond Restrictions.
A Review on Temporal Data Mining
-201-
In the medical field, the clinical temporal data is used in the healthcare system where the digital system-
based medical report has generated by applying the data mining process within less time and effort. Here,
a more advanced automated expert system has needed which takes automated input data from the user end
and generatesa report as an output to the user directly with the prescriptions. That kind of system will be
helpful for human society with financial as well as health aspects because in health care time matters and
it is also helpful for the human daily life routine.
The same report generation task performs in all domains through the data mining process. So, there it
needful to make it fully automated for getting results within less time and effort.
4.5 Recommended Systems
Recommended System is the data-driven system, used for the business domain [99].User’s Time-
dependent data fetch and stored as historical data which are getting from the browser history, social
media, and different other platforms by the service providers. Google used its customer’s data to provide
services to them and also make a profit from the same data in different ways. The recommender system
facilitates the user as well as a service provider for better results in less effort.
The system performs the task with a single click but it also has an issue related to human error which
directly impacts the security of the user’s data. The challenge of continuous filtration of temporal data
based on their authenticity, reliability, and quality parameters can be resolved through apply data
engineering and feature engineering to such systems (Service providers/recommended system).
This review finds the problems at the system level in the data mining process. Any manual process-based
system required more time and effort. So, the process should be automated which consumes less time and
effort with more accurate output. The various researchers have developed a manual and also a semi or
fully automated system for data mining task performance for better results within less effort and
time[100][71][89][94][97][101][102][103]. The automated system can execute all instructions
automatically in digital form, the machine can learn like a human and make the decision accordingly, and
performs better results with more accuracy. Some automated systems are available for data mining tasks
[97][101][102][103]. A fully automated system for the prediction of supervised learning has been
developed by Singh et al. [104] who presented the new automated method for the prediction of supervised
learning in which the prediction methodology integrated with four phases: data preprocessing conducted,
computing the missing value for the cleaning process, the prediction type selection phase and the last
phase for the best model selection for the prediction. The fully automated process has been used in the
proposed method, which does not require domain knowledge and gives the best results.
Overall, the temporal data mining-based applications and related problems at the system level have been
discussed and concluded that the manual systems consume more time in the mining process for
information retrieval. So, the automated data mining applications will be the future of information
Retrieval with less time and effort in all domains.
In the future, research work can be extended to build a more compact and reliable automated system to
solve the problem at the system level for Data Mining in different domains such as (Healthcare,
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-202-
Agriculture, Education, Automobiles, Stock exchanges, etc.).Because the advanced world believes in and
requires plug-and-play systems with fast execution and fewer efforts. So, this is very needful to explore
expert systems in all domains which are helpful to fulfill the human requirements in less time and effort.
5. DISCUSSION
The temporal data mining process is helpful for better decision-making systems in different fields.
This survey covers the importance of the temporal data mining with three leveled structures of the whole
methodology that is data level as preprocessing of an input and model level as a modeling and the System
consists of the integration of data preprocessing and modeling process for temporal data mining. The
summarization of issues and challenges as findings at each level of the temporal data mining process in
our study is as follows:
• Temporal dependency is the main challenge at an initial level of the data mining process which affects
the methodology of the data mining process. Data preprocessing is also challenging here because the
temporal data decomposition depends on temporal data components (levels, trends, seasonality, and
residuals). Here, a sequence data pattern exists that indicates, it’s not easy to disturb and manipulate
the data.
• The challenging task in temporal data mining is the feature selection and extraction, dimensional
reduction, and missing value imputation in temporal datasets because of sequential time-dependency.
It should be done at the data level computation for better results.
• It is very difficult to handle and compute a huge amount of data. It can be solved with different
sampling techniques or by applying advanced machine learning techniques concerning data set size.
• Hard to know about the appropriate model for the computation of temporal data for a different
purpose. Various factors of temporal data and the model itself matter for the same. The different
approach-based models perform differently and give different results on the same temporal dataset.
• The data mining task perform by the manual as well as the machine. The system also performs slowly
in case of a huge amount of data or complex model structure. Another challenge at the system level is
the human interruption in between the mining process which should also affect the computation speed.
• Temporal data mining should be carried past and present data in the computation process to predict
future information. So, Memory consumption is another issue in the temporal data mining process.
• Automation of data mining needed to require minimum human efforts but there is a problem that it
requires updating the domain-specific data to learn and auto knowledge enhancement by self-learning.
The above-mentioned problems can be solved with different methodologies and techniques as the
extension of this paper. In future work, we will try to resolve them efficiently by implementing more
advanced deep learning models and other approaches with providing the relevant outcomes.
A Review on Temporal Data Mining
-203-
6. CONCLUSION
This review provides the detailed temporal data mining and categorized the problems of temporal data
mining into three leveled structures i.e. data, model, and system. This review provides the issues and
challenges based on the experimental study for the temporal data mining that are temporal properties of
datasets at the data level, appropriate method selection at the model level, and time-consuming manual
system at the system level. It shows the problems related to data mining require continuous work from
researchers’ side in temporal data mining.
Conflict of Interest
The authors declare that they have no conflict of interest.
Acknowledgement
I am thankful to cited authors and their research work.
REFERENCES
1. J. Han, J. Pei and M. Kamber, Data mining: concepts and techniques, Elsevier, 2011.
2. H. Kaur, H. S. Pannu and A. K. Malhi, "A systematic review on imbalanced data challenges in machine
learning: Applications and solutions," ACM Computing Surveys (CSUR), vol. 52, p. 1-36, 2019.
3. M. K. Gupta and P. Chandra, "A comprehensive survey of data mining," International Journal of Information
Technology, p. 1-15, 2020.
4. J. Verbraeken, M. Wolting, J. Katzy, J. Kloppenburg, T. Verbelen and J. S. Rellermeyer, "A Survey on
Distributed Machine Learning," ACM Computing Surveys (CSUR), vol. 53, p. 1-33, 2020.
5. A. Purwar and S. K. Singh, "Issues in data mining: A comprehensive survey," in 2014 IEEE International
Conference on Computational Intelligence and Computing Research, 2014.
6. S.-H. Liao, P.-H. Chu and P.-Y. Hsiao, "Data mining techniques and applications-A decade review from
2000 to 2011," Expert systems with applications, vol. 39, p. 11303-11311, 2012.
7. C. C. Aggarwal and S. Y. Philip, "Data mining techniques for associations, clustering and classification," in
Pacific-Asia Conference on Knowledge Discovery and Data Mining, 1999.
8. F. Coenen, "Data mining: past, present and future," Knowledge Engineering Review, vol. 26, p. 25-29, 2011.
9. M. S. B. PhridviRaj and C. V. GuruRao, "Data mining-past, present and future-a typical survey on data
streams," Procedia Technology, vol. 12, p. 255-263, 2014.
10. P.-T. Yang, K.-H. Yang, C.-C. Chen and S.-M. Horng, "Subjective Association Rule Mining: From Point-
based Ranking Sequence to Interval-based Temporal Sequence," in Proceedings of the 2018 10th
International Conference on Machine Learning and Computing, 2018.
11. A. K. Jain, M. N. Murty and P. J. Flynn, "Data clustering: a review," ACM computing surveys (CSUR), vol.
31, p. 264-323, 1999.
12. V. Chandola, A. Banerjee and V. Kumar, "Anomaly detection: A survey," ACM computing surveys (CSUR),
vol. 41, p. 1-58, 2009.
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-204-
13. D. Xu and Y. Tian, "A comprehensive survey of clustering algorithms," Annals of Data Science, vol. 2, p.
165-193, 2015.
14. M. Gupta, J. Gao, C. C. Aggarwal and J. Han, "Outlier detection for temporal data: A survey," IEEE
Transactions on Knowledge and data Engineering, vol. 26, p. 2250-2267, 2013.
15. Q. Zhao and S. S. Bhowmick, "Association rule mining: A survey," Nanyang Technological University,
Singapore, p. 135, 2003.
16. L. A. Kurgan and P. Musilek, "A survey of knowledge discovery and data mining process models," The
Knowledge Engineering Review, vol. 21, p. 1-24, 2006.
17. A. R. Post and J. H. Harrison Jr, "Temporal data mining," Clinics in Laboratory Medicine, vol. 28, p. 83-
100, 2008.
18. L. Sevilla-Lara, S. Zha, Z. Yan, V. Goswami, M. Feiszli and L. Torresani, "Only Time Can Tell: Discovering
Temporal Data for Temporal Modeling," arXiv preprint arXiv:1907.08340, 2019.
19. A. Bertone, "A Matter of Time: Machine Learning and Temporal Data Mining," 2007.
20. S. Laxman and P. S. Sastry, "A survey of temporal data mining," Sadhana, vol. 31, p. 173-198, 2006.
21. R. Campos, G. Dias, A. M. Jorge and A. Jatowt, "Survey of temporal information retrieval and related
applications," ACM Computing Surveys (CSUR), vol. 47, p. 1-41, 2014.
22. M. M. Gaber, A. Zaslavsky and S. Krishnaswamy, "Mining data streams: a review," ACM Sigmod Record,
vol. 34, p. 18-26, 2005.
23. P. Roy, G. Perez, J.-C. Régin, A. Papadopoulos, F. Pachet and M. Marchini, "Enforcing structure on
temporal sequences: the allen constraint," in International conference on principles and practice of
constraint programming, 2016.
24. L. D. Golagani, N. Nelaturi and S. R. Kurapati, "Deep neural network-based approach for processing
sequential data," CSI Transactions on ICT, vol. 8, p. 263-270, 2020.
25. S. Knight, A. F. Wise and B. Chen, "Time for change: Why learning analytics needs temporal analysis,"
Journal of Learning Analytics, vol. 4, p. 7-17, 2017.
26. J. F. Roddick and M. Spiliopoulou, "A survey of temporal knowledge discovery paradigms and methods,"
IEEE Transactions on Knowledge and data engineering, vol. 14, p. 750-767, 2002.
27. T. Mitsa, Temporal data mining, CRC Press, 2010.
28. A. Chapman, E. Simperl, L. Koesten, G. Konstantinidis, L.-D. Ibáñez, E. Kacprzak and P. Groth, "Dataset
search: a survey," The VLDB Journal, vol. 29, p. 251-272, 2020.
29. P. Chaudhari, D. P. Rana, R. G. Mehta, N. J. Mistry and M. M. Raghuwanshi, "Discretization of temporal
data: a survey," arXiv preprint arXiv:1402.4283, 2014.
30. A. Kusiak, "Data Farming Methods for Temporal Data Mining," Intelligent Systems Laboratory, vol. 2139, p.
52242-1527, 2005.
31. G. Ozsoyoglu and R. T. Snodgrass, "Temporal and real-time databases: A survey," IEEE Transactions on
Knowledge and Data Engineering, vol. 7, p. 513-532, 1995.
A Review on Temporal Data Mining
-205-
32. V. Radhakrishna, P. V. Kumar and V. Janaki, "A survey on temporal databases and data mining," in
Proceedings of the The International Conference on Engineering & MIS 2015, 2015.
33. DatasetSource, Shampoo sale dataset, https://machinelearningmastery.com/time-series-datasets-for-
machine-learning/, 2021.
34. DatasetSource, Minimum daily temperature, https://machinelearningmastery.com/time-series-datasets-for-
machine-learning/, 2021.
35. DatasetSource, Monthly sunspot dataset, https://machinelearningmastery.com/time-series-datasets-for-
machine-learning/, UCI MachineLearninig Repository, 2021.
36. DatasetSource, Daily female birth dataset, https://machinelearningmastery.com/time-series-datasets-for-
machine-learning/, 2021.
37. Dataset-Source, EEG-Eye_State Dataset, https://archive.ics.uci.edu/ml/datasets/EEG+Eye+State, UCI
Machine Learning Repository, 2021.
38. Dataset-Source, Occupancy Detection Dataset,
https://archive.ics.uci.edu/ml/datasets/Occupancy+Detection+, UCI Machine Learning Repository, 2021.
39. Dataset Source, Ozone layer Dataset, UCI Machine Learning Repository, https://archive.
ics.uci.edu/ml/datasets/ozone+level+detection, 2021.
40. DatasetSource, Wine quality dataset, UCI Machine Learning Repository, https://archive.ics.
uci.edu/ml/datasets/wine+quality, 2021.
41. DatasetSource, Credit card default dataset, UCI Machine Learning Repository, https://archive.ics.
uci.edu/ml/datasets/default+of+credit+card+clients, 2021.
42. Dataset-Source, London Merged Bike Sharing Dataset, https://www.kaggle.com/hmavrodiev/london-bike-
sharing-dataset, Kaggle, 2021.
43. DatasetSource, Gracious dataset, Kaggle, https://www.kaggle.com/heeraldedhia/groceries-dataset, 2021.
44. DatasetSource, Sale transaction dataset, UCI Machine Learning Repository,
https://archive.ics.uci.edu/ml/datasets/Sales_Transactions_Dataset_Weekly, 2021.
45. DatasetSource, Bread basket Dataset, Kaggle, https://www.kaggle.com/mittalvasu95/the-bread-basket, 2021.
46. DatasetSource, Online retail dataset, UCI Machine Learning Repository,
https://archive.ics.uci.edu/ml/datasets/Sales_Transactions_Dataset_Weekly, 2021.
47. DatasetSource, Boston crime dataset, Boston government, https://data.boston.gov/dataset/crime-incident-
reports-august-2015-to-date-source-new-system, 2021.
48. DatasetSource, House hold power consumption, Kaggle, https://www.kaggle.com/uciml/electric-power-
consumption-data-set, 2021.
49. DatasetSource, Airline Passenger dataset, Machine learning mastery, https://github.com
/jbrownlee/Datasets/blob/master/airline-passengers.csv, 2021.
50. DatasetSource, Air quality dataset, UCI Machine Learning Repository,
https://archive.ics.uci.edu/ml/datasets/Air+Quality, 2021.
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-206-
51. Dataset-Source, WeatherAUS Dataset, https://www.kaggle.com/jsphyg/weather-dataset-rattle-package,
Kaggle, 2021.
52. DatasetSource, Bike Sharing dataset, UCI Machine Learning Repository, https://archive.
ics.uci.edu/ml/datasets/bike+sharing+dataset, 2021.
53. DatasetSource, KC house pricing dataset, Data world, https://data.world/adv34715/housing-price-and-
population-change, 2021.
54. Dataset-Source, Gold Price Forecasting, https://www.kaggle.com/arashnic/learn-time-series-forecasting-
from-gold-price, Kaggle, 2021.
55. Z. Xing, J. Pei and E. Keogh, "A brief survey on sequence classification," ACM Sigkdd Explorations
Newsletter, vol. 12, p. 40-48, 2010.
56. S. Aghabozorgi, A. S. Shirkhorshidi and T. Y. Wah, "Time-series clustering-a decade review," Information
Systems, vol. 53, p. 16-38, 2015.
57. P. A. Gutiérrez, M. Perez-Ortiz, J. Sanchez-Monedero, F. Fernandez-Navarro and C. Hervas-Martinez,
"Ordinal regression methods: survey and experimental study," IEEE Transactions on Knowledge and Data
Engineering, vol. 28, p. 127-146, 2015.
58. C. Kamath, "On mining scientific datasets," in Data Mining for Scientific and Engineering Applications,
Springer, 2001, p. 1-21.
59. R. Rawassizadeh, E. Momeni, C. Dobbins, J. Gharibshah and M. Pazzani, "Scalable daily human behavioral
pattern mining from multivariate temporal data," IEEE Transactions on Knowledge and Data Engineering,
vol. 28, p. 3098-3112, 2016.
60. P. Fournier-Viger, J. C.-W. Lin, R. U. Kiran, Y. S. Koh and R. Thomas, "A survey of sequential pattern
mining," Data Science and Pattern Recognition, vol. 1, p. 54-77, 2017.
61. M. Ghorbani and M. Abessi, "A new methodology for mining frequent itemsets on temporal data," IEEE
Transactions on Engineering Management, vol. 64, p. 566-573, 2017.
62. C. Bettini, X. S. Wang, S. Jajodia and L. Jia-Ling, "Discovering temporal relationships with multiple
granularities in time sequences," IEEE Transations on Knowledge and Data Engineering, vol. 10, 1998.
63. C. Bettini, X. S. Wang, S. Jajodia and others, "Mining temporal relationships with multiple granularities in
time sequences," IEEE Data Eng. Bull., vol. 21, p. 32-38, 1998.
64. J. W. Lee, Y. J. Lee, H. K. Kim, B. H. Hwang and K. H. Ryu, "Discovering temporal relation rules mining
from interval data," in Eurasian Conference on Information and Communication Technology, 2002.
65. Y. J. Lee, J. W. Lee, D. J. Chai, B. H. Hwang and K. H. Ryu, "Mining temporal interval relational rules from
temporal data," Journal of Systems and Software, vol. 82, p. 155-167, 2009.
66. Y.-C. Chen, W.-C. Peng and S.-Y. Lee, "Mining temporal patterns in time interval-based data," IEEE
Transactions on Knowledge and Data Engineering, vol. 27, p. 3318-3331, 2015.
67. T.-c. Fu, "A review on time series data mining," Engineering Applications of Artificial Intelligence, vol. 24, p.
164-181, 2011.
68. P. Esling and C. Agon, "Time-series data mining," ACM Computing Surveys (CSUR), vol. 45, p. 1-34, 2012.
A Review on Temporal Data Mining
-207-
69. A. Baheti and D. Toshniwal, "Trend analysis of time series data using data mining techniques," in 2014 IEEE
International Congress on Big Data, 2014.
70. H. Wang, X. Ding, J. Li and H. Gao, "Rule-based entity resolution on database with hidden temporal
information," IEEE Transactions on Knowledge and Data Engineering, vol. 30, p. 2199-2212, 2018.
71. N. Günnemann-Gholizadeh, "Machine Learning Methods for Detecting Rare Events in Temporal Data,"
2018.
72. S. A. Mirroshandel and G. Ghassem-Sani, "Towards unsupervised learning of temporal relations between
events," Journal of Artificial Intelligence Research, vol. 45, p. 125-163, 2012.
73. S. Ram??rez-Gallego, B. Krawczyk, S. Garc??a, M. Wo?niak and F. Herrera, "A survey on data
preprocessing for data stream mining: Current status and future directions," Neurocomputing, vol. 239, p.
39-57, 2017.
74. M. Goebel and L. Gruenwald, "A survey of data mining and knowledge discovery software tools," ACM
SIGKDD explorations newsletter, vol. 1, p. 20-33, 1999.
75. A. Douzal-Chouakria, J. A. Vilar and P.-F. Marteau, Advanced Analysis and Learning on Temporal Data,
Springer, 2016.
76. A. Douzal and A. M. A. LIG, "Machine Learning on temporal data".
77. S. G. West and J. T. Hepworth, "Statistical issues in the study of temporal data: Daily experiences," Journal
of personality, vol. 59, p. 609-662, 1991.
78. T. G. Dietterich, "Machine learning for sequential data: A review," in Joint IAPR international workshops on
statistical techniques in pattern recognition (SPR) and structural and syntactic pattern recognition (SSPR),
2002.
79. G. Nguyen, S. Dlugolinsky, M. Bobák, V. Tran, Á. L. Garc??a, I. Heredia, P. Mal??k and L. Hluch?,
"Machine Learning and Deep Learning frameworks and libraries for large-scale data mining: a survey,"
Artificial Intelligence Review, vol. 52, p. 77-124, 2019.
80. J. S. Yoo and S. Shekhar, "Similarity-profiled temporal association mining," IEEE Transactions on
Knowledge and Data Engineering, vol. 21, p. 1147-1161, 2008.
81. J. Ni, B. Cao, B. Yao, P. Yu and L. Li, "ARTAR: Temporal association rule mining algorithm based on
attribute reduction," in 2016 First IEEE International Conference on Computer Communication and the
Internet (ICCCI), 2016.
82. A. Savasere, E. R. Omiecinski and S. B. Navathe, "An efficient algorithm for mining association rules in large
databases," 1995.
83. J. Hipp, U. Güntzer and G. Nakhaeizadeh, "Algorithms for association rule mining_a general survey and
comparison," ACM sigkdd explorations newsletter, vol. 2, p. 58-64, 2000.
84. J. S. Yoo, "Temporal data mining: similarity-profiled association pattern," in Data mining: foundations and
intelligent paradigms, Springer, 2012, p. 29-47.
85. E. Winarko and J. F. Roddick, "Discovering richer temporal association rules from interval-based data," in
International Conference on Data Warehousing and Knowledge Discovery, 2005.
AJMI 15(2), July-Dec, 2023 Rashi Jaiswal, Brijendra Singh
-208-
86. J.-H. Jeon and M.-S. Kim, "A study of criterion for efficient clustering estimation of temporal data," The
Journal of The Institute of Internet, Broadcasting and Communication, vol. 11, p. 139-144, 2011.
87. T. W. Liao, "Clustering of time series data_a survey," Pattern recognition, vol. 38, p. 1857-1874, 2005.
88. W. Lin, M. A. Orgun and G. J. Williams, "Temporal data mining using hidden markov-local polynomial
models," in Pacific-Asia Conference on Knowledge Discovery and Data Mining, 2001.
89. D. Alberg, M. Last, R. Neuman and A. Sharon, "Induction of mean output prediction trees from continuous
temporal meteorological data," in 2009 IEEE International Conference on Data Mining Workshops, 2009.
90. S. Laine and T. Aila, "Temporal ensembling for semi-supervised learning," arXiv preprint arXiv:1610.02242,
2016.
91. M. H. Saraee and B. Theodoulidis, "Knowledge discovery in temporal databases," 1995.
92. O. Sagi and L. Rokach, "Ensemble learning: A survey," Wiley Interdisciplinary Reviews: Data Mining and
Knowledge Discovery, vol. 8, p. e1249, 2018.
93. Z. Karevan and J. A. K. Suykens, "Transductive LSTM for time-series prediction: An application to weather
forecasting," Neural Networks, vol. 125, p. 1-9, 2020.
94. R. Kalla, S. Murikinjeri and R. Abbaiah, "An Improved Demand Forecasting with Limited Historical Sales
Data," in 2020 International Conference on Computer Communication and Informatics (ICCCI), 2020.
95. P. M. R. S. M. R. S. K. M. C. K. a. S. A. N. Shirodkar, "Credit Card Fraud Detection Techniques - A Survey,"
in 2020 International Conference on Emerging Trends in Information Technology and Engineering (ic-
ETITE), Vellore, India,, 2020.
96. H.-P. Kriegel, K. M. Borgwardt, P. Kröger, A. Pryakhin, M. Schubert and A. Zimek, "Future trends in data
mining," Data Mining and Knowledge Discovery, vol. 15, p. 87-97, 2007.
97. R. Maiocchi, B. Pernici and F. Barbic, "Automatic deduction of temporal information," ACM Transactions
on Database Systems (TODS), vol. 17, p. 647-688, 1992.
98. F. Wang, N. Lee, J. Hu, J. Sun, S. Ebadollahi and A. F. Laine, "A framework for mining signatures from
event sequences and its applications in healthcare data," IEEE transactions on pattern analysis and machine
intelligence, vol. 35, p. 272-285, 2012.
99. A. Belhadi, Y. Djenouri, J. C.-W. Lin and A. Cano, "A Data-Driven Approach for Twitter Hashtag
Recommendation," IEEE Access, vol. 8, p. 79182-79191, 2020.
100. G. P. Zhang, "Time series forecasting using a hybrid ARIMA and neural network model," Neurocomputing,
vol. 50, p. 159-175, 2003.
101. A. Kaul, S. Maheshwary and V. Pudi, "Autolearn_Automated feature generation and selection," in 2017
IEEE International Conference on data mining (ICDM), 2017.
102. M. Feurer, A. Klein, K. Eggensperger, J. Springenberg, M. Blum and F. Hutter, "Efficient and robust
automated machine learning," in Advances in neural information processing systems, 2015.
103. P. Kerschke, H. H. Hoos, F. Neumann and H. Trautmann, "Automated algorithm selection: Survey and
perspectives," Evolutionary computation, vol. 27, p. 3-45, 2019.
104. B. Singh and R. Jaiswal, "Automation of Prediction Method for Supervised Learninig," in Confluence 2021,
Noida, 2021.