Learning analytics using deep learning techniques for efficiently managing educational institutes

Abstract

Increasing numbers of higher education institutions see themselves as service providers, catering primarily to the needs of its students. The improvement of student performance is a top priority for universities. It is critical to first assess the present situation of the students before designing a program to improve their performance. Higher education administrators face a significant problem in predicting a student's future success. The goal of this study is to learn what factors influence college students' decision on a major. It will be possible to forecast students' behavior, attitudes, and performance with the use of predictive tools and procedures. Predicting student performance ahead of time makes it possible to take proactive measures to raise achievement levels. To obtain a high education standard, several attempts have been made to forecast student performance. However the accuracy of these predictions falls short of the desired level of excellence. Machine learning approaches including Artificial Neural Network, Nave Bayes, and SVM are being studied. A University Data Set from UCI Machinery is used in the experimental investigation.

Full text

Learning analytics using deep learning techniques for efficiently
managing educational institutes
Ravi Kishore Veluri
a
, Indrajit Patra
b
, Mohd Naved
c
, Veduri Veera Prasad
a
, Myla M. Arcinas
d
,
Shehab Mohamed Beram
e
, Abhishek Raghuvanshi
f
a
Aditya Engineering College(A), Surampalem, India
b
An Independent Researcher, PhD from NIT Durgapur, West Bengal, India
c
Department of Business Analytics, Jagannath University, Delhi-NCR, India
d
Associate Professor, Behavioral Science Department, De La Salle University, Philippines
e
Research Scholar, Department of Computing and Information Systems, Sunway University, Malaysia
f
Department of Computer Engineering, Mahakal Institute of Technology, Ujjain, India
article info
Article history:
Available online xxxx
Keywords:
Learning analytics
Deep learning
Educational data mining
Machine learning
Student performance
Classification
Prediction
abstract
Increasing numbers of higher education institutions see themselves as service providers, catering primar-
ily to the needs of its students. The improvement of student performance is a top priority for universities.
It is critical to first assess the present situation of the students before designing a program to improve
their performance. Higher education administrators face a significant problem in predicting a student’s
future success. The goal of this study is to learn what factors influence college students’ decision on a
major. It will be possible to forecast students’ behavior, attitudes, and performance with the use of pre-
dictive tools and procedures. Predicting student performance ahead of time makes it possible to take
proactive measures to raise achievement levels. To obtain a high education standard, several attempts
have been made to forecast student performance. However the accuracy of these predictions falls short
of the desired level of excellence. Machine learning approaches including Artificial Neural Network, Nave
Bayes, and SVM are being studied. A University Data Set from UCI Machinery is used in the experimental
investigation.
Copyright Ó2022 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Confer-
ence on Advances in Materials Science
1. Introduction
Data mining [1 2] aids companies in discovering and under-
standing hidden patterns in large collections by leveraging their
existing reporting skills. These patterns are then included into data
mining algorithms, which are then used to correctly forecast indi-
vidual behavior. In light of this knowledge, organizations are better
able to allocate resources and employees. By using data mining to
anticipate how many students will enroll in a course ahead of time,
an institution can take proactive measures before a student drops
out. Data mining can also help an institution in better allocation of
resources by accurately predicting the probable number of stu-
dents in a particular course.
This study examines data mining’s capabilities and potential
applications in higher education. Data mining uses a combination
of explicit knowledge, powerful analytical capabilities, and subject
experience to uncover hidden trends and patterns. Prediction mod-
els [345] build on these trends and patterns to make new observa-
tions based on the present data. To discover new patterns, trends,
and correlations in large amounts of data stored in repositories, a
method utilizing pattern recognition technology and statistical
and mathematical methodologies is required. In order to do data
mining on very big or raw datasets, either supervised or unsuper-
vised data mining methods should be used. It’s important to
remember that no data mining can be done without interacting
with unitary data first. Machine learning is used in many real
world areas [67].
EDM [89] converts raw data into useful facts and knowledge for
various educational contexts by using it as input. This information
may assist educational policymakers, school administrators,
instructors, and students in making well-informed decisions about
how to manage and utilize educational resources. With data-
driven decision-making, current educational practices and learning
resources can be improved. As educational information systems
https://doi.org/10.1016/j.matpr.2021.11.416
2214-7853/Copyright Ó2022 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Conference on Advances in Materials Science
Materials Today: Proceedings xxx (xxxx) xxx
Contents lists available at ScienceDirect
Materials Today: Proceedings
journal homepage: www.elsevier.com/locate/matpr
Please cite this article as: Ravi Kishore Veluri, I. Patra, M. Naved et al., Learning analytics using deep learning techniques for efficiently managing educa-
tional institutesgiven names and surnames to make sure that we have identified them correctly and that they are presented in the desired order. Carefully

(EIS) have evolved, a vast amount of student data has been avail-
able. This demonstrates the necessity of employing EDM to exam-
ine the learning habits of pupils. EDM aids in the accurate
evaluation of educational institutions in order to get the most
out of the available learning resources.
EDM makes use of research data to identify educational prob-
lems and propose solutions. It tries to look for patterns in curricu-
lum, learning behavior, and student family data from various
educational institutions that have yet to be uncovered. With the
EDM, we hope to gain a better understanding of the current causes
and effects of schooling.
2. Investigation of machine learning and deep learning
techniques for analytics and prediction
Predictive models are developed using classification and regres-
sion techniques in supervised learning. Support vector machines
(SVM), naive Bayes classification approaches, decision trees, closest
neighbor, logistic regression, discriminate analysis, and neural net-
works are examples of traditional classification algorithms.
To anticipate the outcome of unlabeled datasets, unsupervised
learning is used. Clustering is the most used unsupervised learning
technique. Some simple clustering approaches include the K-
means clustering algorithm, K-means clustering algorithms, hier-
archical clustering, and hidden Markova models. There are numer-
ous supervised and unsupervised learning algorithms; however
their application differs greatly from one circumstance to the next.
As a result, choosing the right machine learning algorithm yields
superior results in prediction and classification procedures. Choos-
ing the correct algorithm, on the other hand, is frequently a chal-
lenging undertaking. As a result, the article was thoroughly
examined in order to determine the one that best suited the clas-
sification algorithm for the heart disease prediction system. Classi-
fication is the procedure in predictive analysis that accurately
classifies the provided input data and maps it to their correspond-
ing classes. There are two sorts of data: labeled data and unlabeled
data. The labeled data contains a large number of predictor quali-
ties as well as a single target attribute. The class label is denoted by
each value of the target characteristics. The predictor attributes are
the only ones in the unlabeled attributes. The primary goal of the
classification process is to accurately predict the class of unlabeled
data using classification models developed from labeled instances
(historical data). To begin, a training model is constructed for
which the corresponding class (or target values) is known. The
training data model provides a summary of the relationship
between the data components. This training data model is used
to forecast target values when the target values are unknown.
The predicted values are then compared to the known values or
labeled data to determine the classification process’s accuracy. This
method is known as data model testing, and the data utilized for
testing is referred to as test data or evaluation data. It assesses
the predictability of the process [9].
A root node, branches, and leaf nodes are all present in each
decision tree. The root node is at the top, while the remaining
nodes are leaf or branch nodes. The decision rule or test is imposed
by the internal node on one or more properties of the provided
data. The output is defined by the branch node. Decision trees
are a well-known classification approach since they do not require
any prior knowledge of data distribution. Furthermore, it performs
well with noisy and ambiguous data.
J. Ross Quinlan, a researcher, developed the ID-3 algorithm
(Iterative Dichotomiser-3), which is the first evolved decision
tree-based system. This algorithm is based on entropy and infor-
mation gain measurements. The original dataset starts with a base
nodule and computes the entropy measure of the functional char-
acteristics for each iteration. The attribute with the lowest error
rate (entropy) and most information gain is chosen as a split attri-
butes, and the dataset is split to generate the subset of attributes
based on it. Unless the algorithm is accurately classified to its tar-
get classes, it is recursively repeated on every subset of data. The
decision tree is built with a non terminal node, and the terminal
nodes are defined by the branch’s final subset. The non terminal
node is defined by the split attribute, while the class labels are rep-
resented by the terminal node. It employs the ID-3-based decision
tree model developed to efficiently classify and predict heart prob-
lems at an early stage. Using well-known decision tree techniques
such as CART and ID-3, a prediction model with a large health data-
set is developed. For validation, a 10-fold cross validation approach
is utilized. The results show that utilizing decision tree classifica-
tion approaches, an accurate and efficient model of prediction
models may be developed. The ID-3 algorithm gives better out-
comes with less datasets and better computation measures. How-
ever, when dealing with continuous and massive data sets, such as
electronic health records, it becomes computationally expensive,
lowering the performance metric. Another important disadvantage
of ID-3-based decision tree classification algorithms is model over
fitting, which gradually reduces the accuracy of the health data
categorization process [10].
Author in [11] describe the decision tree classification model.
First, a classification model is created that accurately classifies data
instances. It employs a top-down approach to categorization, mov-
ing from the root node to the leaf nodes. It chooses an efficient
attribute and divides the given data into a subset of datasets based
on entropy measures. For the decision-making process, the attri-
bute with the highest normalized information gain is used. It is
computationally efficient since it prunes branches with unneces-
sary features. It also handles missing values and properties with
numeric values well. When dealing with numeric qualities, the
decision tree becomes more difficult.
According to [12], the decision tree C-5 is an enhanced version
of the C4.5 method that was designed to alleviate the limitations of
the C4.5 algorithm. It allows for faster computing tasks while
requiring less store space and tree size. It allows for boosting and
automatically removes qualities that aren’t needed for the classifi-
cation process. It employs a search constraint over the training
dataset. An independent test set is used to validate the final classi-
fication results. The use of association rules efficiently links the
cardiac risk factor to illness severity assessments. The C-5 decision
tree methodology decreases the number of association rules while
increasing accuracy.
In his article [13] describe that classification and regression tree
is a well-known method of decision tree based classifier. Every
base nodule represents a distinct input as well as distinct base
points across the variable. It is assumed that the input attribute
value is a single digit. The output variable, represented by the leaf
node, is used for prediction. It is based on discriminate analysis and
creates a statistical model to categories the dataset with greater
accuracy. It is effective on both categorical and continuous attri-
butes. In his research, [14] proposed that an ensemble learning
approach conducts classification and regression operations. During
the training phase, it builds a large number of decision trees and
uses regression methods to predict the outcomes of the individual
trees. It has a low variance and quickly links the various aspects of
the given data for prediction purposes. The reason for the initial
lack of commitment to this technique is that the random forest
classification algorithms are difficult to interpret.
Authors [13] explain in their study that information processing
is carried out by highly linked neurons. This method is widely used,
and its main applications include pattern recognition and data
classification. It creates the network by connecting nodes, which
are referred to as neurons. The process of signal transduction from
Ravi Kishore Veluri, I. Patra, M. Naved et al. Materials Today: Proceedings xxx (xxxx) xxx
2

one neuron to another is accomplished through the usage of con-
necting nodes. The artificial neural networks’ input signals are real
numbers, while the output units are nonlinear intakes. The
weighted edge size gradually raises or decreases the signal
strength at the associated edges. At the nodes, a preset threshold
value is set, and neurons can only send signals if they are greater
than or equal to the fixed threshold value. Artificial neurons are
typically depicted in a layered fashion. Each layer applies its own
set of transformations to the inputs provided. Signals often move
from the first to the last layer by traversing the middle layers
numerous times. The primary notion behind artificial neural net-
work approaches is that they are used to solve issues in the same
way that human brains do. Medical diagnosis, video and image
identification are only a few of the important uses of artificial neu-
ral networks.
A K-nearest neighbor technique [11] is the most robust algo-
rithm utilized in pattern recognition and data categorization oper-
ations. The distance functions or similarity measure are the
fundamental concept underlying K-nearest neighbor algorithms.
It saves the state of all instances and uses the similarity metric
to classify freshly defined instances. For efficient categorization
procedures, it employs the instance-based learning method. A
new instance of the dataset is categorized depending on the major-
ity of votes cast by its adjacent classes. For both the training and
test datasets, the distance measure is computed. The algorithm’s
initial step is to select a value for k and calculate the distance
between the instances using the k value.
SVM is a binary linear classification algorithm that is not prob-
abilistic. It creates a training model that categorizes the samples
into one or more target classes. The data objects are represented
in space as points. The objects of different categories are separated
by a visible gap, causing its width to expand. The target classes of
the new instances are mapped based on which side of the gap they
land on. When the input datasets are not labeled, the support vec-
tor machine also supports non-linear classification. Because there
are no target classes to which the instances can be mapped, the
support vector machine uses an unsupervised learning approach
to categories the data. After clusters are built based on functions,
new instances are added to them. The paper presents an effective
model-based recommendation system based on non-linear sup-
port vector machine [15]. Non-linear support vector machine tech-
niques are the most extensively used methodology for dealing with
unlabeled data, and they are employed in a variety of industrial
applications.
3. Framework for performing learning analytics of student data
Fig. 1 depicts a system for student data categorization and per-
formance prediction. A student data collection is utilized as input
in this approach. The data set has been preprocessed. The data his-
tograms are equalized in conjunction with wavelet de noising. The
primary advantage of this approach is that it not only equalizes the
histogram but also adjusts for information loss. Principal compo-
nent analysis is used to extract features. The classification step’s
goal is to identify the student’s category on the basis of analytics
Preprocessing of Data Set
by Histogram Equalization
Input Data Set
Classification
SVM,
Naive Bayes
ANN
Feature Extraction using
PCA
Prediction of Student
Performance
Classification
Result
Fig. 1. Framework for Student Performance Classification and Prediction.
75
80
85
90
95
100
ANN SVM Naïve Bayes
Accuracy in %
Accuracy in %
Fig. 2. Accuracy Results of Classification Algorithms for University Data Set.
Ravi Kishore Veluri, I. Patra, M. Naved et al. Materials Today: Proceedings xxx (xxxx) xxx
3

of features supplied in the data set. In machine learning, there are
several classifiers available, including Naive Bayes, Support Vector
Machine (SVM), and Artificial neural network (ANN).
4. Results
University data set [16] is used for experimental study. This
data set consists of 285 instances. This data set contains seventeen
attributes. The accuracy of classification achieved by different
machine learning algorithms is shown below in Fig. 2.
5. Conclusion
Predicting a student’s future success is a huge challenge for
higher education administration. The purpose of this research is
to discover what factors influence college students’ major selec-
tion. With the application of predictive tools and techniques, it will
be feasible to forecast students’ behavior, attitudes, and perfor-
mance. Predicting student performance in advance allows for
proactive efforts to enhance achievement levels. Several attempts
have been made to forecast student performance in order to
achieve a high education standard. The accuracy of these forecasts,
however, falls short of the anticipated level of excellence. This
paper offered a framework based on machine learning for doing
learning analytics on university student data. The supplied data
collection is also subjected to data preparation. Machine learning
methods such as Artificial Neural Network, Nave Bayes, and SVM
are being investigated. The experimental inquiry makes use of a
University Data Set from UCI Machinery. The experimental results
have proved that the accuracy of artificial neural network method
is better as far as classification of student data is concerned.
Declaration of Competing Interest
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared
to influence the work reported in this paper.
References
[1] L. Ji, X. Zhang, L. Zhang, Research on the Algorithm of Education Data Mining
Based on Big Data, in: 2020 IEEE 2nd International Conference on Computer
Science and Educational Informatization (CSEI), 2020, pp. 344–350, https://doi.
org/10.1109/CSEI50228.2020.9142529.
[2] A. Aleem, M.M. Gore, Educational Data Mining Methods: A Survey, in: 2020
IEEE 9th International Conference on Communication Systems and Network
Technologies (CSNT), 2020, pp. 182–188, https://doi.org/10.1109/
CSNT48778.2020.9115734.
[3] A. Hicham, A. Jeghal, A. Sabri, H. Tairi, A Survey on Educational Data Mining
[2014-2019], International Conference on Intelligent Systems and Computer
Vision (ISCV) 2020 (2020) 1–6, https://doi.org/10.1109/
ISCV49265.2020.9204013.
[4] S. Kovalev, A. Kolodenkova, E. Muntyan, Educational Data Mining: Current
Problems and Solutions, V International Conference on Information
Technologies in Engineering Education (Inforino) 2020 (2020) 1–5, https://
doi.org/10.1109/Inforino48376.2020.9111699.
[5] N. Khodeir, Student Modeling Using Educational Data Mining Techniques, in:
2019 6th International Conference on Advanced Control Circuits and Systems
(ACCS) & 2019 5th International Conference on New Paradigms in Electronics
& information Technology (PEIT), 2019, pp. 7–14, https://doi.org/10.1109/
ACCS-PEIT48329.2019.9062874.
[6] Ravi Manne, Snigdha Kantheti, Sneha Kantheti, Classification of Skin cancer
using deep learning, Convolutional Neural Networks - Opportunities and
vulnerabilities- A systematic Review, Int. J. Modern Trends Sci. Technol., ISSN:
2455-3778, 06 (11) (2020) 101–108. https://doi.org/10.46501/IJMTST061118.
[7] R. Manne, Machine learning techniques in drug discovery and development,
Int. J. Appl. Res. 7 (4) (2021) 21–28.
[8] M.M. Arcinas, A Blockchain Based Framework for Securing Students
Educational Data, Linguistica Antverpiensia 2 (2021) 4475–4484.
[9] Myla M. Arcinas, Guna Sekhar Sajja, Shazia Asif, Sanjeev Gour, Ethelbert
Okoronkwo, The Role of Data Mining in Education for Improving Students
Performance for Social Change, Turkish J. Physiotherapy Rehabil., ISSN 2651-
4451, 32 (3) 6519–6526.
[10] S. Shrestha, M. Pokharel, Machine Learning algorithm in educational data,
Artif. Intell. Transform. Busin. Soc. (AITB) 2019 (2019) 1–11, https://doi.org/
10.1109/AITB48515.2019.8947443.
[11] B. Hssina, A. Merbouha, H. Ezzikouri, M. Erritali, A comparative study of
decision tree ID3 and C4.5, Int. J. Adv. Comp. Sci. Appl. 4 (2) (2014), https://doi.
org/10.14569/issn.2156-557010.14569/SpecialIssue.2014.040203.
[12] L.M. Crivei, G. Czibula, G. Ciubotariu, M. Dindelegan, Unsupervised learning
based mining of academic data sets for students’ performance analysis, in:
2020 IEEE 14th International Symposium on Applied Computational
Intelligence and Informatics (SACI), 2020, pp. 000011–000016, https://doi.
org/10.1109/SACI49304.2020.9118835.
[13] B. Al Breiki, N. Zaki, E.A. Mohamed, Using Educational Data Mining Techniques
to Predict Student Performance, in: 2019 International Conference on
Electrical and Computing Technologies and Applications (ICECTA), 2019, pp.
1–5, https://doi.org/10.1109/ICECTA48151.2019.8959676.
[14] T.O. Olaleye, O.R. Vincent, A Predictive Model for Students’ Performance and
Risk Level Indicators Using Machine Learning, 2020 International Conference
in Mathematics, Computer Engineering and Computer Science (ICMCECS),
2020, pp. 1–7, http://doi.org/10.1109/ICMCECS47690.2020.240897.
[15] X. Yang, M. Li, Y. Zhang, J. Ning, Cost-sensitive naive bayes classification of
uncertain data, J. Sci. World 9 (8) (2014) 1897–1904.
[16] http://archive.ics.uci.edu/ml/datasets/university.
Ravi Kishore Veluri, I. Patra, M. Naved et al. Materials Today: Proceedings xxx (xxxx) xxx
4