International Journal of Computer Applications Technology and Research

Volume 8–Issue 05, 157-160, 2019, ISSN:-2319–8656

Student Performance Prediction

Isha D Shetty Dipshi Shetty Sneha Roundhal

Department of Information Department of Information Department of Information
Technology Technology Technology
VESIT VESIT VESIT
Chembur, India Chembur, India Chembur, India

Abstract:Predicting Students performance beforehand can be very beneficial for educational institutions to improve their
teaching quality. This paper proposes to predict students performance by considering their academic details. For
classification, decision tree, Random forest, Support vector machine, Linear regression, Naive Bayes are used. This paper
also focuses on the use of Artificial Neural network algorithms such as MLP and Convolutional Neural network for
predicting the performance.

Keywords: Artificial Neural Network, Performance Prediction, Convolutional Neural Network, Multilayer perceptron, SMOTE,
SGPI,CGP.I

after their Higher Secondary education) and Direct Second

1. INTRODUCTION
Today every educational institution handles and deals with Year (DSE-who join after completing their diploma degree).
large amount of student data which can be beneficial for a The SGPI(Student Grade Performance Index) of a particular
number of reasons. One of the important application of such student varies from .01 to .99 and can be difficult to analyze
data is predicting student performance. Such a prediction can and predict, hence to get efficient prediction the academic
be useful not only for the students but also for teachers/mentors. performance data is categorized.
Mentors can provide special assistance to the students who are It has 6050 instances and 55 attributes. The dataset contains
on the verge of failing. In order to determine which category a students Personal details like parents name, date of birth,
student lies, such data can be quite helpful. This application can address etc and academic features like marks of all semesters,
be used by any prominent school or colleges. It can be used to HSC percentages, SSC percentages etc. There is a mix of the
predict the pointer ranges or percentage range for future numeric and nominal attribute.
semester exams. These ranges can be predicted using a number
of data mining algorithms such as classification algorithms, Some of the attributes considered while training the model are
rule-based algorithms, ensemble methods, and neural :
networks. The main aim of this project is the selection of Table -1: Dataset
features that show a strong relationship with a target attribute
that is to be predicted from a high dimensional dataset. We have Attributes
evaluated and compared the number of algorithms such as
decision tree, random forest, support vector machine, naive
Bayes and neural networks by applying them on the dataset The Student Type (Regular or DSE)
rest of the paper provides an explanation on nature of neural
networks along with the results of our evaluation.
SSC Board
1.1 Objective
Today every educational institution handles and deals with SSC Percentage
large amount of student data which can be beneficial for a
number of reasons. One of the important application of such
data is predicting student perfo HSC Board
2. OVERVIEW OF DATASET
This dataset includes actual academic details of the students of HSC Percentage
one of the prominent engineering institute of Mumbai
University. Engineering students of mumbai university are
classified as Regular students (who join engineering colleges

www.ijcat.com 157
 International Journal of Computer Applications Technology and Research
Volume 8–Issue 05, 157-160, 2019, ISSN:-2319–8656

DSE Board
Fig -1: Result of correlation matrix using heatmap

DSE Percentage

Semester Marks (Graduation)

Number of KT’s

Education Gap

Branch

Admission Type(CAP or Minority) The features such as 12_Result,10_Result,Sem1,Sem2, Sem3,

Sem4 were selected.Thus, the dataset was trained using these
features. This is applicable for the dataset of both regular and
Diploma students.
3. DATA PREPROCESSING
The dataset is prepared to apply data mining techniques. 4. METHODOLOGY
Preprocessing methods such as data cleaning, transformation of 3 category
variables, data balancing and data partitioning had to be The student’s SGPI is converted into 3 categories. These
applied. categories are:
There are 17 numeric attributes with null values. To deal with if sgpi<=4 the 0
null values, semester pointers and 10th percentage columns are If sgpi<=7 then 1
replaced with their respective column medians. On the other else 2
hand, the columns such as the number of KTs were replaced The categories for semester wise SGPI are predicted by
with 0 as in the dataset there is proper mention of whether the applying ensemble algorithms -Random forest, Bagging and
person has a kt. So in case if the value of this attribute is NAN, Boosting, rule-based algorithms- Naive Bayes, classification
then it means the student has no KT. The SSC percentages algorithms - decision tree and artificial neural networks-
which are in terms of CGPI were converted into percentages. Convolutional neural network and Multilayer Perceptron.
The dataset was split into the training and testing data. The ratio
of the training and testing data is 8: 2. 5 category
The student’s SGPI is converted into 3 categories. These
3.1 Data Balancing
Unbalanced data typically refers to a problem with categories are:
classification problems where the classes are not represented if sgpi<=4 the 0
equally. The unbalanced data is solved using SMOTE also If sgpi<=5 then 1
called as synthetic sampling. In both 3 and 5 category section, If sgpi<=6 then 2
the SGPI range<=4 i.e class 0 has less number of instances If sgpi<=7 then 3
comparatively. Since this class is important to identify weak else 4
students, it cannot be neglected. This technique is similar to The categories for semester wise s are predicted by applying
oversampling where the instances of class 0 are increased to ensemble algorithms -Random forest, Bagging and Boosting,
handle imbalance. rule-based algorithms- Naive Bayes, classification algorithms -
decision tree and artificial neural networks- Convolutional
3.2 Feature Selection neural network and Multilayer Perceptron.
The attributes such as name,email-id, phone-number, mother’s
name, father’s name, address are unique and are dropped as it
shows the least relationship with the target attribute. The
Semester pointers and 12th board results and 10th board results
showed a linear relationship with the target variable.
Correlation states how the features are related to each other or
the target variable.Thus, the correlation matrix was calculated
for finding the relationship. For instance, the semester 5 SGPI
range was to be predicted. So, it showed a strong relationship
with the following features.

www.ijcat.com 158
 International Journal of Computer Applications Technology and Research
Volume 8–Issue 05, 157-160, 2019, ISSN:-2319–8656

5. RESULTS
For Regular Students
The results of these algorithms were compared and the results
are:

Table -1: Results of Algorithms

Algorithms Accuracy

3 Category(*100) 5 Category(*100)

Random Forest 0.99 0.99

Gaussian Naive 0.78 0.87

Bayes For Diploma Students
The results of these algorithms were compared and the results
Support vector 0.70 0.81 are:
machine
Table -1: Results of Algorithm

Bagging 0.99 0.99

Algorithms Accuracy
Boosting 0.98 0.99

Decision Tree 0.98 0.98 3 Category(*100) 5 Category(*100)

Linear Regression 0.73 0.78

Random Forest 0.99 0.99
MLP 0.85 0.77
Gaussian Naive 0.97 0.95
Bayes
CNN 0.92 0.80
Support vector 0.93 0.88
Chart -1: For 3 Category machine

Bagging 0.99 0.99

Boosting 0.98 0.99

Decision Tree 0.98 0.99

Linear Regression 0.73 0.79

MLP 0.62 0.77

CNN 0.88 0.78

Chart -1: For 3 Category

Chart -2: For 5 Category

www.ijcat.com 159
 International Journal of Computer Applications Technology and Research
Volume 8–Issue 05, 157-160, 2019, ISSN:-2319–8656

Mr.Vinayak Hegde Department of Computer Science

Amrita Vishwa Vidyapeetham University, Mysuru
Campus -2016 International Conference on Data Mining
and Advanced Computing.
[2] Proposed academic students’ performance prediction
model: A Malaysian case study.
[3] Performance Prediction of Engineering students using
Decision trees .-R. R. Kabra, R. S. Bichkar -International
Journal of Computer Application.
[4] Student performance analysis using classification
techniques -International Journal of Pure and Applied
Mathematics
[5] Data Mining Techniques for Predicting Student
Chart -2: For 5 Category Performance -2015 IEEE International Conference on
Engineering and Technology (ICETECH), 20th March
2015, Coimbatore, TN, India
[6] GritNet: Student Performance Prediction with Deep
Learning -2018 Educational Data Mining
Organisation,Mountain View , California
[7] Students' Performance Prediction Using Deep Neural
Network -International Journal of Applied Engineering
Research ,Volume 13, Number 2 (2018)
[8] Student Performance Prediction using Machine Learning
-International Journal of Engineering Research &
Technology (IJERT) ISSN: 2278-0181
IJERTV4IS030127 www.ijert.org (This work is licensed
under a Creative Commons Attribution 4.0 International
License.) Vol. 4 Issue 03, March-2015
[9] Early Prediction of Students Performance using Machine
Learning Techniques-International Journal of Computer
Applications (0975 – 8887) Volume 107 – No. 1,
6. CONCLUSION December 2014
In this paper, the classification, rule-based learning, ensemble [10] Using Machine Learning to Predict Student Performance-
methods, and neural network based algorithms are employed in University of Tampere Faculty of Natural Sciences
student information to predict the students' division on the Software Development M. Sc. Thesis Supervisor: Jorma
premise of previous information. The accuracy was calculated Laurikkala,June 17
using a confusion matrix. The confusion matrix proved to be a
[11] Machine Learning Approach to Predict Student Academic
good metric for unbalanced data.
Ensemble methods work best with such structured data as Performance-International Journal for Research in
compared to its counterpart i.e neural networks. The neural Applied Science & Engineering Technology (IJRASET)
networks require a large amount of data for training.Since the ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor:
dataset used is less complex and is in a structured format, 6.887 Volume 6 Issue IV, April 2018
ensemble methods work best as compared to Convolutional [12] Predicting Students’ Academic Performance in the
neural networks and Multilayer perceptron for both 3 University Using Meta Decision Tree Classifiers -Authors
categories and 5 categories prediction.This study may be quite : Shanthini, A., G. Vinodhini and R.M. Chandrasekaran
helpful for both students and teachers for improving the [13] Machine Learning Based Student Grade Prediction: A
performance and improving the future results of the weak Case Study -Authors : Zafar Iqbal , Junaid Qadir, Adnan
students. Noor Mian and Faisal Kamiran

7. ACKNOWLEDGEMENT
Our sincere thanks to Dr.Mrs. Shalu Chopra ,HOD of
Department of Information Technology ,VESIT for providing
the data about academic details of our college students .We are
grateful to her for helping and guiding us throughout the
project.

8. REFERENCES
[1] Prediction of student performance using educational
mining by Ms.Tismy Devasia, Ms.Vinushree T,

www.ijcat.com 160