You are on page 1of 38

A Machine Learning Assisted Web Based Student

Performance Monitoring System

by

Examination Roll: 210197

A research project report submitted to the


Institute of Information Technology
in partial fulfillment of the requirements for the degree of
Master of Science in
Information Technology

Supervisor: Md. Biplob Hosen


Lecturer, IIT, JU

Institute of Information Technology


Jahangirnagar University
Savar, Dhaka-1342
May 2023
DECLARATION

I hereby declare that the outcome of this project is based on my own findings.
Other works are mentioned as a source of information. This report has never been
submitted for a degree before, in whole or in part.

MD.Uthba Bin Anowar Binoy


Roll:210197

ii
CERTIFICATE

It is to verify that the project named A Machine Learning Assisted Web


Based Student Performance Monitoring System has been prepared and sub-
mitted by MD. Uthba Bin Anowar Binoy in sectional fulfilment of the demand
for the degree of Master of Science (M.Sc.) in Information and Communication Tech-
nology on May 16, 2023.

Md. Biplob Hosen


Supervisor
Lecturer, IIT,JU

Accepted and approved in partial fulfilment of the requirement for the degree
Masters of Science (MSc.) in ICT.

Dr. M. Shamim Kaiser Dr. Risala Tasin Khan Md. Fazlul Karim Patwary
Chairman Member Member

Prof. Dr. Md. Saiful Islam


External Member
IICT, BUET

iii
ACKNOWLEDGEMENTS

I am lucky to be able to express myself and thanks to those who all together took
part in this process of making report.
I have performed my project by the supervision of Md.Biplob Hosen, Lecturer, IIT,
Jahangirnagar University. In our work he has helped me in many ways. He helped
me with giving books, journals. I may not have done this project in time without his
help. I am pleased with his guidance , proposition, inspiration and cordial company.
I thank Md. Biplob Hosen, Lecturer, IIT, Jahangirnagar University for his advice
he gave me so I can complete my project in time. Beside that, I am also pleased to
people of IIT.
I am also thankful to those source which helped me in completing my work.

iv
ABSTRACT

In our modern day, the education sector is quite competitive. Many pupils are
unable to stay on course until the finish. Some students are unaware of their weak-
nesses. Consequently, they are unable to enhance their total performance.The aim
of this project is to develop a machine learning based web application, where stu-
dents can predict their CGPA on a particular course by filling up some details on
their profile. The system uses a previously trained dataset of over 1000 entries taken
from university students to predict the CGPA. Additionally, teachers can also view
the students’ progress and the classes they are taking. The system is built using a
Python-based machine learning algorithm that utilizes TensorFlow to make predic-
tions. The model is then saved in .js format and reloaded into Node.js using the
@tensorflow/tfjs npm package. Once a student fills out their profile, the backend
predicts their CGPA and returns the results to the frontend.The system is designed
to help students understand how much effort they need to put into their courses to
achieve their desired results. It is also useful for teachers to monitor their students’
progress and identify any areas where additional support may be required.

Keywords: Students, Performance,Prediction,Machine learning,Deep Learning,Data


Mining

v
LIST OF FIGURES

Figure

3.1 Tech Stack Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 15


3.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Relationship Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1 MAE loss and MSE of ANN and CNN for train and test dataset . . 24

vi
TABLE OF CONTENTS

DECLARATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

CERTIFICATE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . iv

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

LIST OF ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . vi

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

CHAPTER

I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Limitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Research Outline . . . . . . . . . . . . . . . . . . . . . . . . . 3

II. Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Analysis of related Work . . . . . . . . . . . . . . . . . . . . 4


2.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

III. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.1 Dataset and Data Collection . . . . . . . . . . . . . . . . . . 8


3.2 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Classifier Design . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4.1 Machine Learning . . . . . . . . . . . . . . . . . . . 10
3.4.2 Deep Learning . . . . . . . . . . . . . . . . . . . . . 13

vii
3.5 Web Application Design . . . . . . . . . . . . . . . . . . . . . 14
3.5.1 Requirements Analysis . . . . . . . . . . . . . . . . 14
3.5.2 Technical Requirements . . . . . . . . . . . . . . . . 16
3.5.3 Architecture Overview . . . . . . . . . . . . . . . . 16
3.5.4 System Components . . . . . . . . . . . . . . . . . . 17
3.5.5 API Design . . . . . . . . . . . . . . . . . . . . . . 19

IV. Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.1 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . 21


4.2 Stastistical metrics . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.1 MAE . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2 MSE . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.3 RMSE . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.4 R2 Square . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 ML Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.4 DL results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

V. Conclusion and Plan of Future . . . . . . . . . . . . . . . . . . 26

5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2 Plan of Future . . . . . . . . . . . . . . . . . . . . . . . . . . 26

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

viii
CHAPTER I

Introduction

1.1 Background

Using techniques such as data analysis, machine learning algorithms, etc., a stu-
dent performance prediction and monitoring system allow us to track a student’s
performance over time and forecast the student’s performance in a given course or aca-
demic program. This type of system utilizes relevant data such as grades, attendance,
and demographic information, among others, to achieve optimal results.This type of
system is advantageous for both students and instructors.By providing students with a
clear comprehension of their academic progress and a predicted outcome, the student
performance monitoring system can assist them in maintaining their motivation and
focus.In addition, the system is capable of providing students with individualized feed-
back on their performance, assets, and weaknesses. This feedback can assist them in
identifying areas for development and formulating strategies to overcome obstacles.In
addition, the system can help students become more self-aware of their academic per-
formance and assume responsibility for their own learning. This can assist them in
becoming more accountable and engaged students.On the other hand,this type of sys-
tem can help teachers intervene early and provide targeted support to students who
are at risk of falling behind or underperforming academically.The system can provide
instructors with insightful information about their students’ academic performance,
enabling them to make data-driven decisions regarding instructional strategies, lesson
planning, and assessment.In addition, the system can facilitate teacher-student com-
munication by providing a forum for feedback, discussion, and collaboration. This can
help to strengthen teacher-student relationships and create a more supportive learn-
ing environment.By providing a more individualized approach to education, a student
monitoring system can enhance inclusivity in schools. By analysing data to deter-
mine the specific requirements of each student, the system is able to provide targeted

1
interventions and resources to help students succeed. This can be especially bene-
ficial for students from diverse backgrounds or with special needs, who may require
additional assistance or accommodations to succeed. Additionally, the system can
help to level the playing field by identifying students who are falling behind early and
providing them with opportune interventions to help them catch up. This can prevent
pupils from falling behind and from experiencing feelings of exclusion or marginal-
ization. In addition, the system can enhance communication between instructors,
students, and parents by providing regular updates on student progress and oppor-
tunities for feedback and collaboration.Web-based educational systems are intended
to provide educators and students with the tools and resources necessary to facili-
tate instruction and learning. Learning Management Systems (LMS) are the most
prevalent form of web-based educational system, offering features such as course man-
agement, content management, assessment tools, and communication tools. MOOCs
are web-based courses that are accessible at any time and are available to anyone.
Educational social networks are social media platforms designed specifically for edu-
cational purposes that allow students and teachers to communicate, share resources,
and collaborate. Virtual Learning Environments (VLEs) are web-based platforms
that facilitate teaching and learning activities, such as communication, collaboration,
assessment, and resource sharing. The selection of a system is contingent upon the
specific requirements of the educational institution or individual educator, as well as
the objectives of the teaching and learning activities.

1.2 Problem Statement

Existing student performance monitoring systems have limitations that can af-
fect their accuracy and usefulness. Lack of real-time data, narrow focus on academic
performance, limited adaptability to individual student requirements, and lack of
integration with other educational technologies are some of these limitations.This en-
deavor utilizes data that significantly influences a student’s performance.The system
predicts the CGPA using a dataset of over one thousand university student entries
that has been previously trained.Additionally, very few student monitoring systems
are web-based.Thus, the development of a web-based student mentoring system is
intended.

2
1.3 Objective

• To monitor the overall performance of a particular student throughout the


semester

• To predict the final result

• To predict the result of a particular course

• Giving feedback to a student

• To find out the weakness of a student

1.4 Limitation

1. No classification model for dataset

2. various categorization model delivers varied accuracy

3. The characteristics of attributes evolve over time

1.5 Research Outline

Our report is structured as follows: In Chapter II, we provide a literature re-


view of related work, including explanations of the most important terms used in
this thesis-basic concept, as well as examples of the work of others, as well as a
discussion of how our work differs from theirs and what we hope to gain from the
similarities.Chapter III introduces the system model.

3
CHAPTER II

Literature Review

2.1 Analysis of related Work

Hamsa et al.[1] constructed a model to predict a student’s academic performance


Using Decision Tree and Fuzzy Genetic Algorithm. It utilizes internal marks, ses-
sional marks, and admission score as parameters. The resulting prediction model
can be used to determine each student’s subject-specific performance.Current aca-
demic analytics researches on this topic were examined by Asiah et al.[? ] with a
focus on predicting student academic performance. Previous researchers have pro-
posed a number of methods for developing the optimal performance model using
a variety of student data, techniques, algorithms, and instruments.To predict stu-
dent performance,Kavakchieva et al.[2] employs classification algorithms derived from
data mining.It compares the performance of multiple algorithms and concludes that
the accuracy of the decision tree classifier is the highest.This article [3] provides
a thorough analysis and comparison of current supervised machine learning tech-
niques applied to the problem of predicting student exam performance. Artificial
neural networks achieved the highest precision in general.Using data mining tech-
niques, the student constructs a methodology for monitoring and evaluating student
academic performance. It derives performance prediction indicators and implements
a straightforward system for evaluating and monitoring student performance. The
system is founded on diverse data mining techniques and the deployment of machine
learning processes.Using data mining techniques, Ogor et al.[4] proposes a method
for monitoring and evaluating student academic performance.They derives perfor-
mance prediction indicators and implements a straightforward system for evaluating
and monitoring student performance. The system is founded on diverse data mining
techniques and the deployment of machine learning processes. Shah et al.[5] used ma-
chine learning and deep learning techniques to predict the performance of students,

4
as well as fundamental exploratory data analysis to identify numerous correlations
between student performance and psychographic characteristics.Cluster analysis and
k-mean clustering algorithm-based system were described by Oyelade et al.[6] for an-
alyzing student performance. It was carried out for a Nigerian private institution.Al
et al.[7] introduced two models for estimating students’ final test performance. The
paper used the widely available dataset on mathematics performance supplied by the
University of Minho in Portugal. The dataset consisted of 395 data samples. The
findings of empirical investigations showed that the Support Vector Machine per-
formed marginally better than the K-Nearest Neighbor algorithm, with a correlation
coefficient of 0.96 vs 0.95.In order to visualize patterns of student activity within the
online environment and to establish whether or not these log data may be utilized
to predict student academic success, this research[8] provides a unique approach for
processing course log data gathered from Moodle-based blended courses.GritNet, a
new deep learning-based algorithm were proposed by Kim et al[9] for predicting pupil
performance. It is based on bidirectional long short term memory (BLSTM) and
outperforms the conventional logistic-regression-based method.In this paper[10],they
developed a novel machine learning method for predicting student performance in de-
gree programs.The proposed technique has two primary characteristics: 1) a bilayered
structure consisting of multiple base predictors and a cascade of ensemble predictors;
and 2) a data-driven method founded on latent factor models and probabilistic ma-
trix factorization. The performance of the proposed method is superior to that of
benchmark approaches.In order to construct a predictive model for student perfor-
mance,Yacoob et al.[11] employed a supervised data mining.Using the ’Naïve Bayes’
classification algorithm, it is determined that the greatest predictors of outstanding
students are A+ and A in Multivariate Analysis; A+, A, and A- in SAS Programming;
and A, A-, and B+ in ITS 472.Adhatrao et al.[12] discusses a system that can predict
a student’s future performance based on their past performance using concepts of
data mining techniques. It applies the ID3 and C4.5 classification algorithms to a
data set containing information about students, including gender, marks scored in the
board examinations of classes X and XII, marks and rank in entrance examinations,
and first-year results from the previous cohort of students.This paper[13] proposes a
new model for predicting student performance based on data mining techniques and
new data attributes/features known as student behavioral features. These features
pertain to the interaction between the learner and the e-learning administration sys-
tem. Results indicate that there is a significant correlation between the behaviors of
students and their academic achievement.A monitoring model is discussed for student

5
performance by Mathai et al.[14] The model employs responses from questionnaires,
base registers, and teachers. The model is advantageous for both students and the
institution.Thai et al.[15] examines the application of factorization models for predict-
ing student performance. It contends that these models can be used to replace some
existing standardized tests and discusses the advantages of doing so.Existing pupil
performance analysis and monitoring systems were examined by Kekane et al.[16] It
evaluates student performance based on academic achievement, research and inno-
vation, personal growth, and extracurricular activities. The descriptive statistics of
the students and their GPA from the first semester of their undergraduate studies
are used to predict their academic performance in the undergraduate degree program
by Ibrahim et al.[17].Borkar et al.[18] developed significant guidelines to measure
the association between numerous student traits, thereby enhancing their academic
performance. Experiment is conducted via Weka and a real-time data set accessi-
ble on campus.On two distinct student datasets, Zaffar, Maryam, et al. (2018) [19]
analyze the performance of filter feature selection techniques and classification algo-
rithms.This research[20]aims to build mechanisms for monitoring university students’
performance. In addition to making it easy for system users to view the performance
of students in accordance with the data required for the monitoring process.An an-
alytical strategy are proposed for e-learning data management in this research[21] .
Their primary objective is to identify groups of students based on their responses.
They concentrate on Educational Data Mining (EDM) technique. In their work, data
from a PSLC repository English e-learning course were chosen for a case study in the
validation process.Agarwal et al.[22] proposes a model for predicting the academic
achievement of students inside an academic institution. The algorithm utilized is a
technique for machine learning known as Neural Networks.Machine learning (ML)
algorithms were developed by Rai et al.[23] on a university student dataset to pre-
dict student performance. On the basis of the examination of the results, it has
been determined that the random forest (RF) classifier is more accurate than other
methods such as support vector machine (SVM). Sravani, B., Bala, M. M. (2020,
June)[24] employed linear regression, an algorithm for machine learning, to predict the
academic achievement of students.Different algorithms, including KNN, Nave Bayes,
and Decision Tree are applied by Mohammadi et al.[25] to demonstrate how to employ
Data Mining techniques to identify information in raw student data.This study[26]
investigates machine learning techniques to predict the eventual Grade Point Average
of students based on their personal factors (including gender and living location),
university entry scores, gap year, and academic performance in the first and second

6
year.Gokmen et al.[27] provides a novel way for assessing performance based on fuzzy
logic systems. The performance of students in the Control Technique Laboratory of
the Electricity Education Department of the Technical Education Faculty of Marmara
University was evaluated using fuzzy logic and compared to the conventional evalu-
ating method.This study’s[28] overarching goal is to provide a logically sound system
that will enable educators to keep a close eye on their students’ academic progress,
allowing for prompt adjustments to be made to failing teaching strategies.Hidayah,
I., Permanasari, A.E. and Ratwastuti, N., 2013[29]constructed a student classifica-
tion model using the neuro fuzzy concept; a combination of fuzzy’s IF-THEN rules
and neural network’s capacity to learn, thus this technique can learn from the gener-
ated rules to build the most accurate classification model.A fuzzy inference system is
developed and evaluated by Asopa et al.[30] in MATLAB for fuzzy ITS, which will
assist students improve their learning skills.Rao, D.H., Mangalwede, S.R. and Desh-
mukh, V.B., 2017[31] developed a method for evaluating student performance using
the Fuzzy Inference System (FIS) for the Network Analysis (NA) course taken by
Electronics and Communication Engineering students in their third semester. This
paper discusses the significance of Bloom’s levels in learning and developing critical
thinking abilities for the NA course, as well as in building a scoring rubric by aligning
the rubric requirements with Bloom’s Taxonomy levels, which are then fed into the
FIS.

2.2 Summary

People have performed performance prediction based on a single method or several


methods and their particular field, as evidenced by the preceding discussion. They are
considering attributes depending on their purpose and the possibility of their own.We
constructed our model based on the conditions of our environment and attempted to
forecast which traits effect us the most.

7
CHAPTER III

Methodology

3.1 Dataset and Data Collection

The student modules and teacher modules incorporate and track the historical
and education information of each student and teacher. Over one thousand students
filled up some sort of form inorder to enroll into the prediction system.Some fea-
tures were identified which are related to CGPA.They are department,gender,SSC
HSC GPA,parents’ education and job,major illness,attendance,study hour,internet
facilities, group study,sport/cultural involvement,political involvement,hostel stay-
ing,getting any scholarship,self-income,relational status ,communication skill,confidence
level, previous semester result. By feeding these historical data separate prediction
systems were trained to get certain results. The mentioned factors can influence a
student’s CGPA in various ways. The difficulty of a department’s courses can effect
a student’s performance. According to research, gender influences academic perfor-
mance, with females frequently outperforming males.The past academic performance
of a student at the secondary and advanced secondary levels can be used to pre-
dict their future performance in college.A student’s motivation, expectations, and
educational aspirations can be affected by his or her parents’ educational level and
occupation.
Serious health conditions can result in absences, decreased study time, and de-
creased motivation, all of which can have a negative effect on academic performance.
Attendance is crucial for keeping up with course material and assignments. Time
spent researching can have an effect on academic performance. Students can access
resources, communicate with peers and instructors, and complete assignments if they
have access to dependable internet and technology.Collaboration with colleagues can
facilitate learning and comprehension of course content.Participation in extracurric-
ular activities can aid in the development of soft skills such as leadership and col-

8
laboration, but excessive participation can hinder academic performance.Engaging
in politics can be a diversion from academic pursuits. The environment of a hostel
can be conducive to study, but it can also be a source of distractions.Financial aid
can alleviate the burden of paying for education, allowing students to concentrate
on their studies.Working while studying can reduce a student’s available time and
vitality for study.Depending on the circumstances, relationships can provide either
social support or diversions.Communication skills can influence a student’s ability
to communicate effectively with peers and faculty, which can have an effect on their
academic performance.A student’s motivation and willingness to learn can be affected
by confidence.Academic performance in the future can be predicted by past academic
performance.

3.2 Data Processing

Each value associated with categorical data was converted into distinct numbers
to represent the respective classifications.Instead of male-female, gender data is rep-
resented as a binary value between 0 and 1.There are 36 departments.Therefore, it
is represented by the numbers 0-35.Similar to gender, major illness, and cultural
and political engagement, there are two alternatives for hostel stay and relationship
status.Therefore, they are represented as 0 through 1.Attendance In Class, Internet
Access, and Group Study had the three options.Therefore, these are represented by
the numbers 0 through 2.None of the numerical information has changed. After cat-
egorization, numerical data were used to train machine learning and deep learning
algorithms.

3.3 Data Analysis

We discovered a correlation between confidence and cgoa by analyzing cgpa data


from previous years.Students who are relatively self-assured in decision-making, ex-
tracurricular activities, and study have achieved success.Their average grade point
average ranges between 3.25 and 3.90. Additionally, there is a relationship between at-
tendance and cgpa.Students with a high GPA have a high rate of attendance.Students
who are unmarried have a higher attendance rate than those who are married or in a
committed relationship. Weekly study time and grade point average are directly pro-
portional.It has been observed that unmarried students devote more time to studying
than students who are married or in a relationship.

9
3.4 Classifier Design

3.4.1 Machine Learning

3.4.1.1 Linear Regression

Linear regression is a statistical technique used to characterize the relationship


between two continuous variables, one of which is dependent and the other indepen-
dent. The goal of linear regression is to identify the line of best fit that minimizes
the sum of squared distances between observed data points and predicted values of
the dependent variable based on the independent variable. Y = a + bX represents
the line of best fit, where Y is the dependent variable, X is the independent variable,
an is the intercept, and b is the slope. Simple linear regression, in which there is only
one independent variable, and multiple linear regression, in which there are multiple
independent variables, are both applicable applications of linear regression.

3.4.1.2 Robust Regression

Robust regression is an analysis technique that is less sensitive to anomalies and


other types of deviations from the underlying assumptions of linear regression. Ex-
treme data points or outliers can significantly influence traditional linear regression
models, leading to inaccurate estimates of the relationships between variables. Robust
regression techniques employ alternative statistical methods that are less sensitive to
outliers and can generate more precise estimates of the true relationship between
variables. These techniques include weighted least squares, M-estimators, and other
techniques designed to identify and mitigate the effects of outliers and other data
error sources. Robust regression techniques are especially helpful when the data con-
tain significant outliers or other non-normal distributions that can distort the results
of conventional regression analysis.

3.4.1.3 Ridge Regression

The regularization parameter or lambda () determines the penalty term in Ridge


regression. A higher value of enhances shrinkage and reduces model complexity,
whereas a lower value allows coefficients to vary freely and might lead to overfit-
ting.Ridge regression is beneficial when several predictors are highly linked, which
can create instability and inaccurate coefficient estimates in ordinary least squares
regression. Ridge regression reduces collinearity and improves model accuracy and
generalization by reducing coefficients toward zero.

10
3.4.1.4 Lasso Regression

This type of linear regression incorporates a penalty term for the absolute size of
the regression coefficients. This penalty term, known as L1 regularization, is added
to the objective function of ordinary least squares (OLS) that is minimized during
the regression procedure. As a form of feature selection, the L1 regularization term
reduces coefficients towards zero and sets some of them to precisely zero. This can
help reduce the number of predictors in a model and improve its interpretability. This
formula is especially useful when dealing with high-dimensional data, in which only a
subset of the predictors are pertinent to the response variable. The L1 regularization
can aid in identifying and selecting the most significant predictors, resulting in a
model that is more accurate and compact. In contrast to Ridge regression, however,
Lasso regression can produce sparse solutions in which some coefficients are exactly
zero.

3.4.1.5 Elastic Net Regression

Combining the characteristics of Ridge Regression and Lasso Regression, Elas-


tic Net Regression is a type of regression analysis. It is designed to overcome the
limitations of these two methods and obtain greater prediction accuracy in high-
predictor-variable situations. Elastic Net Regression, like Lasso Regression, can con-
duct variable selection by reducing to zero the coefficients of less significant variables.
In contrast to Lasso Regression, Elastic Net Regression includes an additional regular-
ization parameter that regulates the proportion of Ridge and Lasso penalties. This
enables the method to manage situations involving groups of correlated predictor
variables and to select a group of predictors as opposed to just individual variables.

3.4.1.6 Stochastic Gradient Descent

Stochastic Gradient Descent (SGD) is an iterative optimization algorithm utilized


frequently in machine learning to reduce the cost or loss function of a model. It
is a variant of the gradient descent algorithm that adjusts the model’s parameters
at each iteration based on a randomly chosen subset of the training data. This
increases the algorithm’s computational efficiency and enables it to scale to larger
datasets. At each iteration, the algorithm randomly selects a subset of the training
data, called a batch, and uses only the data in the batch to compute the gradient of
the cost function with regard to the model’s parameters. The step size of the update
is then determined by a learning rate hyperparameter that scales the update in the

11
opposite direction of the gradient. This procedure is repeated until convergence,
which is typically characterized by a maximum number of iterations or a minimal
change in the cost function. Stochastic Gradient Descent is beneficial for training
large neural networks and deep learning models when the dataset contains millions or
billions of records. However, the use of small quantities introduces additional noise
into parameter updates, making the optimization process less stable and more erratic
than batch gradient descent.

3.4.1.7 Random Forest Regressor

Random Forest Regressor is a supervised learning algorithm that is a member of


the ensemble method family. It is utilized for regression tasks in which a continuous
value rather than a discrete label is to be predicted. The algorithm operates by con-
structing multiple decision trees during training and outputting the mean prediction
of the individual trees. Each decision tree is built by arbitrarily selecting a subset of
features and training data. The random selection of features and training data helps
reduce overfitting and enhances the model’s generalization performance. During pre-
diction time, the input features are supplied to each ensemble decision tree, which
then generates a prediction. The final forecast is then determined by averaging all the
individual forecasts. Random Forest Regressor is a well-known algorithm for solving
regression problems in machine learning, and it has been shown to be accurate and
tolerant of chaotic, high-dimensional data.

3.4.1.8 Support Vector Machine Regressor

Support Vector Machine (SVM) regression is a supervised learning algorithm used


for regression tasks. Similar to SVM classification, SVM regression attempts to locate
the optimal hyperplane, but it is used to predict a continuous objective variable as
opposed to separating classes. SVM regression is carried out by locating a hyperplane
that minimizes the difference between the predicted and actual values of the objective
variable. The margin is the distance between the hyperplane and the nearest data
points; the SVM regression model attempts to maximize this margin while simulta-
neously minimizing error. Using kernel functions to accommodate nonlinear data is
one of the primary advantages of SVM regression. These functions transform the
input data into a higher-dimensional space in which it is possible to locate a linear
hyperplane that divides the data into two classes.

12
3.4.2 Deep Learning

3.4.2.1 ANN

An artificial neural network (ANN) is a computational model that mimics the


structure and function of biological neural networks, such as those in the brain. It is
made up of interconnected processing elements, also known as neurons, that process
and analyze complex information in concert. ANNs are programmed to learn from
data and enhance their performance over time through a training process. During
training, a set of input data and corresponding target outputs are presented to the
network, and it modifies its internal parameters to minimize the difference between
its predicted outputs and the actual target outputs. Typically, this procedure is car-
ried out using an optimization algorithm, such as gradient descent. A number of
applications, including image recognition, natural language processing, speech recog-
nition, and predictive analytics, make extensive use of ANNs. They can be composed
of multiple layers of neurons, and the network’s architecture and parameters can be
tailored to the application’s specific requirements.

3.4.2.2 RNN

Recurrent Neural Network(RNN) is a form of neural network designed for sequen-


tial data processing, such as time series or natural language. RNNs are distinguished
by their capacity to maintain a "memory" or state that is updated each time a new
input is processed in sequence. This enables the network to learn patterns and de-
pendencies over time, making it particularly useful for language modeling, speech
recognition, and time series prediction. RNNs have a unique connection structure
that incorporates cycles, enabling information to be transmitted from one time step
to the next. This allows them to analyze data sequences of arbitrary length, mak-
ing them suitable for tasks such as text generation, machine translation, and speech
recognition. LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit)
are common variants of RNNs that were developed to resolve some of the limitations
of traditional RNNs, such as the vanishing gradient problem.

3.4.2.3 CNN

Convolutional Neural Network(CNN) is a form of artificial neural network utilized


frequently in image and video recognition tasks. In a CNN, the input image is pro-
cessed by a series of convolutional layers that extract image features using a set of

13
filters. The network adjusts the filter weights in order to minimize a loss function
that quantifies the difference between the predicted and actual output. The output of
the convolutional layers is then sent to one or more fully connected layers, which per-
form classification or regression tasks using the extracted features. CNNs are widely
used in disciplines such as computer vision and deep learning because their use of
convolutional layers makes them effective at recognizing patterns in images.

3.5 Web Application Design

3.5.0.1 Introduction

The purpose of the web app project is to provide a platform for students to
monitor their performance in their current courses, and for teachers to monitor the
performance of their students. It also aims to provide students with a tool to predict
their overall CGPA for a semester, as well as predict course-wise results. The web
app aims to solve the problem of students not having a clear understanding of their
performance in their courses, and not being able to predict their overall CGPA for a
semester. It also aims to solve the problem of teachers not having a clear understand-
ing of their students’ performance, and not being able to predict the average CGPA
of their courses. The web app will be used in the context of educational institutions,
such as colleges or universities, where students and teachers need a platform to mon-
itor and predict their performance in courses. The app will be accessed by students
and teachers via the internet, allowing for easy and convenient access to performance
data.

3.5.1 Requirements Analysis

The process of thinking about requirement analysis and its importance has been
gone through to ensure that the development goes smoothly. By conducting require-
ment analysis, the necessary functional and non-functional requirements of the web
app have been identified and documented. This helps in developing the web app in a
systematic and organized way, and ensures that all stakeholders’ needs and expecta-
tions are met. The analysis also helps in identifying any potential challenges or risks
early on, which can be addressed proactively, saving time and resources in the long
run.

14
3.5.1.1 The functional requirements of my web app

• User authentication and authorization for students and teachers to access the
system

• Course monitoring and performance tracking for students and teachers

• CGPA prediction for students based on their performance data

• Course-wise result prediction for students

• Average CGPA prediction for courses by teachers

• Secure data storage and management

Figure 3.1: Tech Stack Diagram

15
3.5.1.2 The non-functional requirements of my web app

• Performance requirements: The web app must be able to handle a large number
of users and provide fast response times

• Security requirements: User data must be protected and accessible only by


authorized users. Also, the use of an Identity and Access Management system
is planned for future development.

3.5.2 Technical Requirements

• Programming languages: Node.js, Express.js, React, HTML, CSS, and JavaScript

• Libraries and frameworks: jsonwebtoken, Sequelize ORM, Tensorflow.js, Axios,


and primereact

• Database: MySQL

• Tools: XAMPP, Postman for API testing, and cloud solutions planned for future
development

3.5.3 Architecture Overview

Our web application architecture is based on the microservices pattern. This archi-
tecture comprises several independent microservices, such as Student, Teacher, Score
Prediction, and Communication Services. Each microservice has its own implementa-
tion and can be built using different technology stacks and database implementations.
The Student Service allows basic CRUD operations for students, and its data will be
fed into an Artificial Neural Network for CGPA prediction. The Teacher Service
enables teachers to view and monitor student progress with the necessary permis-
sions. The Score Prediction Service predicts students’ CGPA and course-wise results
based on their performance data, utilizing a machine learning model integrated into
the backend. Finally, the Communication Service handles real-time messaging and
notifications between students and teachers.
Microservices architecture is chosen because it offers several advantages over tradi-
tional monolithic architecture. By breaking the application into multiple independent
microservices, we can deploy and develop each service separately, leading to better
scalability and maintainability. Furthermore, each service can be developed in a
different tech stack and can use a different database implementation. Additionally,
this architecture offers better fault isolation as the failure of one microservice will

16
Figure 3.2: System Model

not affect the other services, resulting in better reliability and robustness. The use
of interservice communication between microservices leads to better modularity and
flexibility.
Scalability and maintainability were the two critical factors that influenced our
decision to use the microservices architecture. By breaking the application into in-
dependent microservices, we can scale each service separately based on its demand,
resulting in better overall scalability of the application. Additionally, since each mi-
croservice has its own implementation, we can modify or update a service without
affecting the other services, leading to better maintainability. Moreover, this architec-
ture enables us to adopt the best technology stacks and database implementations for
each service, resulting in better flexibility and adaptability to future changes. Overall,
the microservices architecture provides us with a robust, scalable, and maintainable
architecture that can accommodate future growth and requirements.

3.5.4 System Components

The following list provides a breakdown of the major components in the web app
and an explanation of their purpose and functionality, as well as how they interact
with each other.

17
3.5.4.1 Student service

This component is responsible for handling basic CRUD operations related to stu-
dent data, such as creating, updating, and deleting student information. It also stores
and manages the data required for CGPA prediction and provides this information
to the Score Prediction Service. The Student Service can be accessed by students
themselves as well as teachers with the necessary permissions.

3.5.4.2 Teacher Service

This component is responsible for handling basic CRUD operations related to


teacher data, such as creating, updating, and deleting teacher information. It also
stores and manages the data required for course-wise predictions and provides this
information to the Score Prediction Service. The Teacher Service can be accessed by
teachers with the necessary permissions.

3.5.4.3 Score Prediction Service

This component is responsible for predicting the CGPA and course-wise results of
students based on their performance data. It uses a machine learning model integrated
in the backend for this purpose. The Score Prediction Service interacts with both the
Student Service and the Teacher Service to fetch the required data for prediction.

3.5.4.4 Communication service

This component is responsible for handling communication between students and


teachers, such as messaging and notifications. It uses web sockets to establish real-
time communication between users. The Communication Service interacts with both
the Student Service and the Teacher Service to fetch user information and permissions.

3.5.4.5 Database

This component is responsible for persistently storing the data required by the
other components. Each microservice has its own database implementation, allowing
for greater modularity and flexibility. The databases used by each microservice can
be different, depending on the specific needs of the service.
These components interact with each other through interservice communication.
For example, the Score Prediction Service fetches data from both the Student Service
and the Teacher Service to make predictions. Similarly, the Communication Service

18
Figure 3.3: Relationship Diagram

fetches user information and permissions from the Student Service and the Teacher
Service to facilitate communication. The use of web sockets for real-time communi-
cation between users is another example of how the components interact with each
other. The databases used by each microservice are also an important means of in-
teraction, as they allow each service to store and retrieve the data it needs to perform
its functions.

3.5.5 API Design

The following API design documentation outlines the endpoints and parameters
for the Student Service, Teacher Service, Course Service, and Score Prediction Service
in our web application. The API endpoints provide functionality for user authentica-
tion and authorization, as well as CRUD operations for student and teacher data. The
Course Service provides information on courses, while the Score Prediction Service
allows for the prediction of course-wise results and CGPA based on student perfor-
mance data. This API design aims to provide a comprehensive and secure way for
our users to interact with our web application.

19
3.5.5.1 Student Service
HTTP Method URL Parameter Response
GET /Students All Users
POST /Students/signup Name, Email, Pass- JSON Web Token,New
word, Roll, Registra- User
tion No
POST /Students/Login Email,Password Json Web Token,New
User
PATCH /Students/:id Information to update Updated user
GET /Students/:id Student ID with Au- User info
thorization
DELETE /Students/:id Student ID with Au- No Content
thorization

3.5.5.2 Teacher Service


HTTP URL Parameter Response
GET /teachers All users
POST /teachers/signup Name,Email,Password,Designation Json web token,new
user
POST /teachers/login Email,Password Json web token,new
user
PATCH /teachers/:id information to update update user
GET /teachers/:id teacher id with authorization User info
DELETE /teachers/:id teacher id with authorization No content

3.5.5.3 Course Service


HTTP Method URL Parameter Response
GET /courses All courses
GET /courses/:id course id with authorization course info

3.5.5.4 Score Prediction Service


HTTP Method URL Parameter Response
POST /Predict Name,Email,Password,Designation Predicted Score
GET /Predict-CGPA Student Information Predicted CGPA

20
CHAPTER IV

Experiment

4.1 Experimental setup

Our objective is to predict a student’s CGPA for the full semester and for a
specific course.Total of 22 attributes are used to do that.Cleaning, converting, and
engineering the data allows for the collection and preprocessing of the dataset for
the 22 attributes. The most effective ML algorithm is selected and trained using the
preprocessed dataset after relevant features are determined using feature selection
techniques. To make judgments about the system’s efficacy and suggest areas for
future research, the findings are examined.

4.2 Stastistical metrics

4.2.1 MAE

Mean Absolute Error, or MAE for short, is a statistical metric used to assess the
variation between two continuous variables. It represents the mean of the absolute
discrepancies between the actual data and the predictions.How closely the forecasts
match the actual values is measured by MAE. An improved fit of the model to the
data is shown by a lower MAE. In regression issues where the objective is to predict
a continuous variable, it is frequently used.

4.2.2 MSE

The statistical metric known as MSE—mean squared error—is used to quantify


the discrepancy between expected and actual values. It is the square root of the
average of the discrepancies between the expected and actual values.How closely the
predictions match the actual values is measured by MSE. An improved fit of the

21
model to the data is shown by a lower MSE. It is used in regression issues where the
objective is to predict a continuous variable.

4.2.3 RMSE

Root Mean Squared Error, also known as RMSE, is a statistical metric used to
assess the discrepancy between expected and observed values. It is the average of the
squared discrepancies between the projected values and the actual values, expressed as
a square root.Similar to MSE, RMSE provides a measurement of how well the forecasts
match the actual values. However, RMSE is used over MSE because it is easier to
read and is expressed in the same units as the objective variable. Additionally, RMSE
is a superior metric when the goal is to reduce the impact of large errors because the
square root operation makes it more sensitive to large errors than MSE.

4.2.4 R2 Square

A statistical metric used to assess the goodness of fit of a regression model is


R-squared, commonly referred to as the coefficient of determination. It calculates
the percentage of the dependent variable’s variance that can be accounted for by
the independent variables.Higher numbers suggest a better fit of the model to the
data, and R-squared runs from 0 to 1. A value of 1 denotes that all the variation in
the dependent variable is explained by the model, whereas a value of 0 denotes that
none of the variance in the dependent variable is explained by the model. Regression
analysis frequently uses R-squared to evaluate a model’s performance and benchmark
it against other models.

4.3 ML Results

The present investigation employed diverse machine learning models to identify


CGPA grades and subsequently evaluated their efficacy based on the Mean Absolute
Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and
R-squared (R2) values. The present study involves a comparison of various models,
namely Linear Regression, Robust Regression, Ridge Regression, Lasso Regression,
Elastic Net Regression, Stochastic Gradient Descent, Random Forest Regressor, SVM
Regressor, Artificial Neural Network 1, Artificial Neural Network 2, and Convolutional
Neural Network. The comparison of ML algorithms is given at 4.1.
The findings indicate that the Convolutional Neural Network (CNN) model ex-
hibited superior performance compared to the other models, as evidenced by its lower

22
Table 4.1: Performance Comparison of Machine Learning Models for CGPA Grade
Detection

Model MAE MSE RMSE R2 Square


Linear Regression 1.20E-01 2.40E-02 1.55E-01 -9.56E-03
Robust Regression 6.15E-01 1.08E+01 3.28E+00 -4.52E+02
Ridge Regression 1.20E-01 2.37E-02 1.54E-01 2.94E-03
Lasso Regression 1.20E-01 2.37E-02 1.54E-01 1.93E-03
Elastic Net Regression 1.20E-01 2.37E-02 1.54E-01 1.94E-03
Stochastic Gradient Descent 1.81E+10 1.44E+22 1.20E+11 -6.07E+23
Random Forest Regressor 1.28E-01 2.91E-02 1.71E-01 -2.25E-01
SVM Regressor 1.31E-01 2.92E-02 1.71E-01 -2.26E-01
Artficial Neural Network 1 2.20E-01 6.64E-02 2.58E-01 -1.79E+00
Artficial Neural Network Model 2 0.3418 0.1406 0.3750 -4.9152
Convolutional Neural Network 0.1200 0.0240 0.1549 -0.0088

MAE, MSE, and RMSE values, which were recorded at 0.11999, 0.02399, and 0.15488,
respectively. The coefficient of determination (R2) for the Convolutional Neural Net-
work (CNN) model was observed to be the most proximal to zero, thereby suggesting
that it exhibited the most optimal conformity to the dataset.
The performance of the second Artificial Neural Network Model (ANN 2) was
notable, as evidenced by its Mean Absolute Error (MAE) value of 0.34176, Mean
Squared Error (MSE) value of 0.14065, and Root Mean Squared Error (RMSE) value
of 0.37503. Nevertheless, the ANN 2’s R2 value was found to be negative, which
suggests that the model did not provide a satisfactory fit for the data.
The Linear Regression, Ridge Regression, Lasso Regression, and Elastic Net Re-
gression models exhibited comparable performance, as evidenced by their respective
MAE values of 0.12 and MSE and RMSE values of 0.0237 and 0.154. The R2 coef-
ficients associated with these models were comparatively deficient, signifying inade-
quate conformity with the data.
The Random Forest Regressor and SVM Regressor models exhibited a slightly
inferior performance compared to the aforementioned models, as evidenced by their
respective MAE values of 0.128 and 0.131. The models’ R2 values exhibited negativity,
which suggests inadequate fitting of the models.
The model of Stochastic Gradient Descent exhibited the poorest performance,
as evidenced by its notably elevated MAE, MSE, and RMSE values of 1.81E+10,
1.44E+22, and 1.20E+11, respectively, and a highly negative R2 value. These results
suggest that this particular model was the least appropriate for the task of detecting
CGPA grades.

23
The CNN and ANN 2 models are anticipated to require a lengthier duration for
training and evaluation owing to their intricate architectures, resulting in a higher
runtime. Conversely, the remaining models are projected to exhibit relatively faster
runtimes. The effective duration of the program’s execution is contingent upon several
variables, including the magnitude of the dataset, the hardware employed, and the
particular implementation of the models.
In brief, the findings suggest that the CNN algorithm exhibits the most favourable
outcomes for detecting CGPA grades, as evidenced by its exceptional performance in
relation to various assessment criteria. The selection of a model is contingent upon
the particular demands and limitations of the given problem.

4.4 DL results

(a) MSE of ANN (b) MAE of ANN

(c) MSE of CNN (d) MAE of ANN

Figure 4.1: MAE loss and MSE of ANN and CNN for train and test dataset

Artificial neural networks (ANNs) are widely utilised models for diverse applica-

24
tions, such as the prediction of cumulative grade point average (CGPA). The present
study involved the training of an artificial neural network (ANN) model utilising a
dataset, wherein the model was comprised of four dense layers, batch normalisation,
and dropout. The training of the model was conducted over a period of 500 epochs,
and a comprehensive record of the epoch-wise particulars of the experiment has been
furnished. The experiment employed loss and mean absolute error (MAE) as the
training and validation metrics, respectively.
The findings indicate that the training loss and mean absolute error (MAE) of
the model exhibited a gradual decline with each epoch, suggesting a corresponding
enhancement in the model’s performance as the training progressed. Nonetheless,
the limited availability of data points poses a significant challenge in conducting a
comprehensive evaluation of the model’s efficacy. In contrast, the validation metrics
indicate that the model’s efficacy experienced a marginal enhancement during the
initial epochs, followed by a consistent plateau throughout the subsequent training
iterations.
In general, the artificial neural network (ANN) model that incorporates four dense
layers, batch normalisation, and dropout techniques demonstrates favourable perfor-
mance in predicting cumulative grade point average (CGPA). Further scrutiny is
necessary to evaluate the efficacy of the model, encompassing the incorporation of
supplementary assessment criteria and a more extensive corpus. Moreover, addi-
tional experimentation involving diverse hyperparameters and architectures may be
required to attain superior outcomes.

25
CHAPTER V

Conclusion and Plan of Future

5.1 Conclusion

Using Machine Learning, we intend to design a system that can monitor and
predict performance. If our system receives adequate user attention, we will be able
to collect a sufficient amount of actual data and train our model to perform optimally.
Students will be able to improve their performance with our assistance.

5.2 Plan of Future

In the coming days, we will attempt to achieve the following:

1. enhancing the database

2. constructing a real-time application

26
References

[1] H. Hamsa, S. Indiradevi, and J. J. Kizhakkethottam, “Student academic perfor-


mance prediction model using decision tree and fuzzy genetic algorithm,” Proce-
dia Technology, vol. 25, pp. 326–332, 2016.

[2] D. Kabakchieva, “Student performance prediction by using data mining classifi-


cation algorithms,” International journal of computer science and management
research, vol. 1, no. 4, pp. 686–690, 2012.

[3] N. Tomasevic, N. Gvozdenovic, and S. Vranes, “An overview and comparison of


supervised data mining techniques for student exam performance prediction,”
Computers & education, vol. 143, p. 103676, 2020.

[4] E. N. Ogor, “Student academic performance monitoring and evaluation using


data mining techniques,” in Electronics, robotics and automotive mechanics con-
ference (CERMA 2007), pp. 354–359, IEEE, 2007.

[5] M. B. Shah, M. Kaistha, and Y. Gupta, “Student performance assessment and


prediction system using machine learning,” in 2019 4th International Conference
on Information Systems and Computer Networks (ISCON), pp. 386–390, IEEE,
2019.

[6] O. J. Oyelade, O. O. Oladipupo, and I. C. Obagbuwa, “Application of k means


clustering algorithm for prediction of students academic performance,” arXiv
preprint arXiv:1002.2425, 2010.

[7] H. Al-Shehri, A. Al-Qarni, L. Al-Saati, A. Batoaq, H. Badukhen, S. Alrashed,


J. Alhiyafi, and S. O. Olatunji, “Student performance prediction using support
vector machine and k-nearest neighbor,” in 2017 IEEE 30th Canadian conference
on electrical and computer engineering (CCECE), pp. 1–4, IEEE, 2017.

27
[8] R. C. Raga Jr and J. D. Raga, “Monitoring class activity and predicting student
performance using moodle action log data,” International Journal of Computing
Sciences Research, vol. 1, no. 3, pp. 1–16, 2017.

[9] B.-H. Kim, E. Vizitei, and V. Ganapathi, “Gritnet: Student performance predic-
tion with deep learning,” arXiv preprint arXiv:1804.07405, 2018.

[10] J. Xu, K. H. Moon, and M. Van Der Schaar, “A machine learning approach for
tracking and predicting student performance in degree programs,” IEEE Journal
of Selected Topics in Signal Processing, vol. 11, no. 5, pp. 742–753, 2017.

[11] W. F. W. Yaacob, S. A. M. Nasir, W. F. W. Yaacob, and N. M. Sobri, “Supervised


data mining approach for predicting student performance,” Indones. J. Electr.
Eng. Comput. Sci, vol. 16, no. 3, pp. 1584–1592, 2019.

[12] K. Adhatrao, A. Gaykar, A. Dhawan, R. Jha, and V. Honrao, “Predicting stu-


dents’ performance using id3 and c4. 5 classification algorithms,” arXiv preprint
arXiv:1310.2071, 2013.

[13] E. A. Amrieh, T. Hamtini, and I. Aljarah, “Mining educational data to predict


student’s academic performance using ensemble methods,” International journal
of database theory and application, vol. 9, no. 8, pp. 119–136, 2016.

[14] P. Mathai and K. Sreekumar, “Students performance tracking using bpm classi-
fication modelling,” in 2021 International Conference on Advances in Electrical,
Computing, Communication and Sustainable Technologies (ICAECT), pp. 1–6,
IEEE, 2021.

[15] N. Thai-Nghe, T. Horváth, L. Schmidt-Thieme, et al., “Factorization models for


forecasting student performance.,” in EDM, pp. 11–20, Eindhoven, 2011.

[16] S. Kekane, D. Khairnar, R. Patil, S. Vispute, and N. Gawande, “Automatic


student performance analysis and monitoring,” Int. J. Innovative Res. Comput.
Commun. Eng, 2016.

[17] Z. Ibrahim and D. Rusli, “Predicting students’ academic performance: comparing


artificial neural network, decision tree and linear regression,” in 21st Annual SAS
Malaysia Forum, 5th September, 2007.

[18] S. Borkar and K. Rajeswari, “Predicting students academic performance using


education data mining,” International journal of computer science and mobile
computing, vol. 2, no. 7, pp. 273–279, 2013.

28
[19] M. Zaffar, M. A. Hashmani, K. Savita, and S. S. H. Rizvi, “A study of feature se-
lection algorithms for predicting students academic performance,” International
Journal of Advanced Computer Science and Applications, vol. 9, no. 5, 2018.

[20] A. Nurmoslim, T. Hidayat, S. Khoirunnisa, T. Oktavia, and N. Limantara, “Stu-


dent performance indicator monitoring systems: A case study abc university,”
in 2017 International Conference on Information Management and Technology
(ICIMTech), pp. 88–93, IEEE, 2017.

[21] A. M. De Morais, J. M. Araujo, and E. B. Costa, “Monitoring student perfor-


mance using data clustering and predictive modelling,” in 2014 IEEE frontiers
in education conference (FIE) proceedings, pp. 1–8, IEEE, 2014.

[22] H. Agrawal and H. Mavani, “Student performance prediction using machine


learning,” International Journal of Engineering Research and Technology, vol. 4,
no. 03, pp. 111–113, 2015.

[23] S. Rai, K. A. Shastry, S. Pratap, S. Kishore, P. Mishra, and H. Sanjay, “Machine


learning approach for student academic performance prediction,” in Evolution in
Computational Intelligence, pp. 611–618, Springer, 2021.

[24] B. Sravani and M. M. Bala, “Prediction of student performance using linear


regression,” in 2020 International Conference for Emerging Technology (INCET),
pp. 1–5, IEEE, 2020.

[25] M. Mohammadi, M. Dawodi, W. Tomohisa, and N. Ahmadi, “Comparative study


of supervised learning algorithms for student performance prediction,” in 2019
International Conference on Artificial Intelligence in Information and Commu-
nication (ICAIIC), pp. 124–127, IEEE, 2019.

[26] D. T. Ha, P. T. T. Loan, C. N. Giap, and N. T. L. Huong, “An empirical study


for student academic performance prediction using machine learning techniques,”
International Journal of Computer Science and Information Security (IJCSIS),
vol. 18, no. 3, pp. 21–28, 2020.

[27] G. Gokmen, T. Ç. Akinci, M. Tektaş, N. Onat, G. Kocyigit, and N. Tektaş,


“Evaluation of student performance in laboratory applications using fuzzy logic,”
Procedia-Social and Behavioral Sciences, vol. 2, no. 2, pp. 902–909, 2010.

29
[28] J. A. Rojas, H. E. Espitia, and L. A. Bejarano, “Design and optimization of
a fuzzy logic system for academic performance prediction,” Symmetry, vol. 13,
no. 1, p. 133, 2021.

[29] I. Hidayah, A. E. Permanasari, and N. Ratwastuti, “Student classification for aca-


demic performance prediction using neuro fuzzy in a conventional classroom,” in
2013 international conference on information technology and electrical engineer-
ing (ICITEE), pp. 221–225, IEEE, 2013.

[30] P. Asopa, S. Asopa, N. Joshi, and I. Mathur, “Evaluating student performance


using fuzzy inference system in fuzzy its,” in 2016 International Conference on
Advances in Computing, Communications and Informatics (ICACCI), pp. 1847–
1851, IEEE, 2016.

[31] D. Rao, S. Mangalwede, and V. Deshmukh, “Student performance evaluation


model based on scoring rubric tool for network analysis subject using fuzzy
logic,” in 2017 International Conference on Electrical, Electronics, Communi-
cation, Computer, and Optimization Techniques (ICEECCOT), pp. 1–5, IEEE,
2017.

30

You might also like