International Journal of Trend in Scientific Research and Development (IJTSRD)

Volume 5 Issue 4, May-June 2021 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470

Discovering Student Dropout

Prediction through Deep Learning
Shashikant Karale, Rajani Pawar, Sharvari Pawar, Poonam Sonkamble
Student, Padmabhushan Vasantdada Patil Institute of Technology, Pune, Maharashtra, India

ABSTRACT How to cite this paper: Shashikant Karale

There have been increased incidences of dropout that have been noticed in the | Rajani Pawar | Sharvari Pawar | Poonam
universities in the recent years. These increased reports have been Sonkamble "Discovering Student Dropout
instrumental in introducing the graduation rate of the course completion rate Prediction through Deep Learning"
for majority of universities all over the globe. Dropouts are highly undesirable Published in
and are an indication of some underlying inconsistencies that have been International Journal
plaguing the course since a long time. Therefore, an effective system for the of Trend in Scientific
purpose of prediction of the dropout rate is the need of the hour. To reach Research and
these goals this research article has utilized machine learning approaches. The Development (ijtsrd),
proposed methodology utilizes the K Nearest Neighbor, Fuzzy Artificial Neural ISSN: 2456-6470,
Network and Decision Tree. This approach has been illustrated in utmost Volume-5 | Issue-4, IJTSRD43700
detail in this research article, highlighting the execution of the various June 2021, pp.1549-
important modules of the methodology. The experimentation has been 1553, URL:
performed to achieve the performance of the approach which has yielded www.ijtsrd.com/papers/ijtsrd43700.pdf
highly accurate results.
Copyright © 2021 by author (s) and
KEYWORDS: Fuzzy Artificial Neural Networks, K Nearest Neighbor, Decision Tree, International Journal of Trend in Scientific
Online Courses Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
Commons Attribution
License (CC BY 4.0)
(http: //creativecommons.org/licenses/by/4.0)

INTRODUCTION
Education is an essential and one of the most integral forms interested fields. This in depth education is entrusted form
of development for a human being. The human being is one by the Universities and colleges. The University provides
of the most intellectually capable organisms on this planet. education effectively through the use of lecturers and
With the highest brain to body ratio, a lot of activities and classes. The interested students pursue the relevant course
other important virtues are the direct result of an increased for a period of time stipulated by the University to get
intellect. As the case with having an intellectual power it proficient in their field. This is essential for the student as
needs to be useful and should be honed effectively in a well as the educational institution to provide the valuable
particular way. This honing of the internet is one of the most education to the students. As universities are designed in a
crucial aspects of growing up. As a child grows up over the capitalistic model they need to make profits to keep
years more and more knowledge is accumulated and the providing education to large number of students.
child learn new ways and techniques to become self-
But there has been noticed that there are increase incidences
sustainable and improve his conscience.
of student dropout from the courses over the past few years.
Education plays an important role in improving the These dropouts are highly problematic and can be effectively
conscience of the individual significantly. This improvement detrimental to the university as well as students. The
it is done through providing and effective education to the increased numbers of dropouts are basically an indication of
children starting from a very young age. This development is some irregularities that are forcing the students to dropout.
essential for the overall health mentally and physically of the This process can be ameliorated if the intentions of the
child. As the child grows learn more about the world around students to dropout are effectively visible in time. Therefore
him and is educated by the parents and at school. School is there is a need for an effective system for the purpose of
an important part of a child's life which teaches him a lot of predicting the student dropout rate in a particular course.
things that would help them later in life and currently. The These predictions can be highly valuable and decreasing the
schools are a heaven of knowledge which imparts valuable dropouts and helping the students complete their course or
education to the children effectively improving their lives degree without any hiccups.
over all.
There have been number of related words that have been
After the education provided by the schools these children analyzed in this research article. These researches have been
then effectively are introduced to various different concepts performed to provide resolution for the increasing rates of
and Fields. And the children utilize these fields and guess dropout that are noticed in universities all over the globe.
their interest to pursue education in those particular There has been increased interest of the researchers

@ IJTSRD | Unique Paper ID – IJTSRD43700 | Volume – 5 | Issue – 4 | May-June 2021 Page 1549
 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
introducing this problem and finding the root cause of the methodology on factors such as financial condition, dropping
increased dropout rates. The findings have been effective in year, and gender. They have used feature selection, logistic
understanding that the prediction of a dropout rate if regression, and neural networks to data using data mining
determine for a student it can be effectively useful in techniques Classification and Regression Tree (CART) and
addressing the problems and the concerns of the student. CHAm tree.
This information is valuable in reducing the rate of dropouts
L. Wang explains the emerging wave of artificial intelligence
for the students and improving the rate of successful
and network information platform technology for the
completion for the university. The prediction has also been
Massive Open Online Courses (MOOC). MOOC is a new
useful in understanding the inconsistencies in the course
classroom model that is greatly hit on the traditional
material which can be rectified in time before a lot of
education model.[5] In the proposed paper they
student’s dropout. Therefore the related words analyzed in
implemented a prediction model by using a time-controlled
this approach have been effective in the designing of our
Long Short-Term Memory neural network (ELSTM).[5]Tasks
technique for the purpose of dropout prediction through the
such as learning evaluation, talent training, curriculum
implementation of machine learning approaches. Literature
development, teaching mode have greater benefits. Some
survey paper provides a thorough analysis of the student
many shortcomings and problems need to be solved in
dropout rate prediction and realizes an effective machine
MOOC. The experimental results of the proposed model are
learning implementation for improving the dropout
higher which shows the effectiveness of the proposed paper.
prediction.
N. Shiratori states since the 1990s a serious problem has
In this paper, section 2 is dedicated for the literature review
been are rising in Japan that is dropout rate has been a major
of past work, Section 3 describes the details of the
point of interest by Yomiuri Shimbun. [6] As the semester
developmental procedure of the model. Section 4 evaluates
progresses students lose the interest in course. In
the results through some experiments and finally section 5
preliminary dropout, status distributes the dropout-related
concludes this research article with the traces of the future
state of students by using the logistic regression model. Thus
scope.
the proposed paper implements model to divide students to
Literature Review examine the issue of student dropout rates. Thus the
T. Hasbun narrates in Organization for Economic Co- proposed research students with low motivation in studies
operation and Development countries student dropout is so faculty member to know the reasons for this low
one of the salient problems. [1] Thus for dropout prediction, motivation an early interview.
Educational Data Mining (EDM) analysts are studying
D. Ktoridou narrates that almost half of engineering students
institutional interventions through students. In the proposed
change their field due to teaching and advising insufficiency,
paper the authors have given importance to extracurricular
the complexity of the engineering curriculum.[7] There are
activities from science degrees engineering and business to
six personality types enterprising, conventional, artistic,
predict dropout. The main aim of this paper is to prove that
social, and realistic. The proposed paper shows the similarity
how extracurricular activities can be suitable factors for
between engineering undergraduate students with
dropout prediction. The results of the proposed technique
personality types. This research is done with 113
have shown better performance.
undergraduate students the main reason for leaving
Di Sun states in the last few years there has been a rapid engineering was poor teaching and advising. The paper
growth in open online courses (MOOCs) but they're a serious provides positive effects on both students and Engineering
problem is arising by the MOOC researchers and providers is departments
dropout prediction. [2] The paper presents a solution for
R. Pereira presents the student dropout prediction thus by
students getting dropouts just by analyzing how much
using the data mining techniques disciplinary and
syllabus can be completed by the students in a course. Thus
institutional data of students from the University of Nariño.
for predicting the student dropout recurrent neural network
[8] By using the classification techniques such as decision
(RNN) and learning resource representation layer is used
trees Socioeconomic and academic student dropout profiles
thus the accuracy of the proposed papers is higher than the
were generated. The proposed model can help to develop
traditional machine learning methodology.
policies and strategies for the students to improve the
J. Canas aims to offer an easy and explainable plan of action activities for studies. In recent years there have been many
to recognize dropout-prone and fail-prone students.[3] For types of research made in the field of education by using
practitioners and researchers students who do not complete Data mining techniques.
there higher education or they interrupt their study in
A. Ortigosa states the main aim of the proposed paper is to
between is one of the major concern. The early prediction of
put a stop to students from leaving the university.[9] The
the student’s dropout may help teachers to reduce dropout
proposed paper collected 11,000 students’ data from the five
rates and also suggest new learning materials for the
years by using the C5.0 algorithm which is one of the
students from failing or not completing the course. In the
successful predictive models and SPA i.e Sistema de
proposed paper the researcher has used tree-based
Prediccioen de Abandono dropout prediction system in
classification models to predict dropout-prone and fail-
Spanish. The SPA model was taken into use in 2017 and is
prone students.
currently in use. Thus by using this model 117,000 risk
M. Mustafa describes for developing countries the national scores to detect the dropout risk of more than 5700 students
and international loss of student’s dropout prediction is one it to support student dropout risk prevention.
of the major concerns. The main aim of this paper is to
K. Murakami performs machine learning techniques to
implement an effective dropout prediction system for
predict dropout rates of current and graduate students by
institutes and colleges.[4] To classify the successful from
collecting their data thus the main aim of the paper is to
unsuccessful students they implemented square test

@ IJTSRD | Unique Paper ID – IJTSRD43700 | Volume – 5 | Issue – 4 | May-June 2021 Page 1550
 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
dropout prevention. [10] The university arranges and Step 2: K Nearest Neighbor – The preprocessed list attained
collects data of students to provide enrollment management. in the previous step is utilized as an input in this step of the
They focus on the student dropout and by using data from procedure. This step of the student dropout detection
both current, graduate, and dropout students. Thus by using performs the K Nearest Neighbor clustering. This is
the machine learning algorithm while predicting these performed to classify the dataset given as an input into
dropout rates it is important to investigate unknown items semantic groups to achieve accurate detection of the student
regarding EMIR. dropout in massive online courses. This procedure is
performed through the steps given below.
K. Santos explains for university student’s dropout
prevention by using various educational data mining Distance Evaluation – The Euclidean Distance formula given
techniques. [11] The algorithm such as RF and Decision in the equation 1 below is being used for the purpose of
Trees some algorithms are not successful to get high evaluation of the distance between the training and testing
outcomes due to the large data and a parameter tuning step. dataset attributes. The dataset is provided in the form of a
Thus the proposed paper uses the different supervised list that is given as an input to this step of the procedure. The
learning techniques such as Decision Trees, K-nearest data stored in the rows of this list is used for the purpose of
Neighbor, Support Vector Machines, Naive Bayes, Neural calculating the distance from the labeled testing dataset, and
Networks, and Random Forests to calculate the prediction stored as the row distance at the end of the list. This is
models for reducing school dropout. performed for all the entries in the list iteratively.
R. Lottering describes in the field of education in South ________ (1)
Africa an advanced development has been seen in this
development dropout rate is noticed. [12] As in higher Where,
education in South Africa student numbers grew by 32.8% ED=Euclidian Distance
from 2006 to 2015 and the dropout rate of 17.1% has been
noticed and it can be doubled in the next ten years. Thus ATi=Attribute at index i
proposed systems use data mining techniques and predict ATj= Attribute at index j
dropout goals. The researches of the proposed papers show
a positive approach to the proposed model. Centroid Estimation – The list with the data and the
respective row distances calculated in the previous step are
PROPOSED METHODOLOGY used in this step as an input. The list is sorted in the
ascending order of the Row distance and subjected to
random data point selection. These selected data points are
K in number and achieving the centroid. These data points
are used to determine the row distance of the selected index.
The obtained row distance is used further to achieve the
boundary of the clusters.
The row distances achieved previously are used to form the
average row distance of the entire list. These average row
distances along with the extracted row distance of the
centroid are used to determine the boundaries of the
Figure 1: Overview of the Proposed Idea clusters in the cluster formation procedure given below.
The proposed methodology of the learning outcomes of Cluster Formation – The centroids and the average row
massive online courses is depicted in figure 1 and it is distance acquired previously are provided as an input to this
explained through following steps step for cluster formation. This step then calculates the
Step 1: Preprocessing and Labeling – The proposed system boundaries of the clusters by addition and subtraction of the
uses the dataset of massive online courses for the purpose of average row distance and the centroid row distance to attain
experimentation. The dataset is downloaded from the URL : the maximum and minimum values respectively. The data is
https://dataverse.harvard.edu/dataset.xhtml?persistentId= then subjected to these boundaries to form the clusters
doi:10.7910/DVN/29779. which are aggregated into a cluster list. This cluster list
sorted into a descending order and the top 3 clusters are
Once Testing dataset of the user is received, it is statically provided to the next step as an input.
stored for the instance learning outcomes. And then
proposed model uses the Training set where it reads and Step 3: Fuzzy Artificial Neural Network – This is one of the
stored in a double dimension list, which is subjected to most important steps in the proposed methodology wherein
preprocessing thereafter. This double dimension list of a double dimensional list is utilized to effectively store the
dataset belongs to training data is divided into certified and attributes of the student details as well as the clusters of
non-certified set based on the Boolean value present in the these values. The fuzzy classification is achieved on this list
column of the dataset. to effectively segregate the clusters and effectively identify
the difference between the maximum and the minimum
Then this dataset is used to label them with the unique cluster values. The difference achieved is then effectively
integer numerical value using hash set functions. After this divided into 5 equal parts to achieve effective fuzzy
all the required rows of the preprocessed list are assigned classification labels. These labels correspond to the fuzzy
with the integer numerical value which is then subjected to crisp values that are segregated as very high medium-low
clustering process in the next step of the model. and very low. The student ID and the score sum of different
activities are stored in a double list to generate the fuzzy
crisp ranges as mentioned in the below algorithm1.

@ IJTSRD | Unique Paper ID – IJTSRD43700 | Volume – 5 | Issue – 4 | May-June 2021 Page 1551
 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
________________________________________________ achieved in the Java programming language. The laptop
utilized for the development of this prediction methodology
ALGORITHM 1: Fuzzy Crisp Value generation
is equipped with an Intel core i3 processor which is
__________________________________________________ supplemented by 500 GB of storage and 6 GB of RAM.
//Input : Student Score list SLST For the purpose of achieving the performance metrics of the
proposed methodology the precision and recall parameters
//Output: FCLST ( Fuzzy Crisp List)
have been assigned. The extraction of the performance of
fuuzyCrispValueGeneration(SLST) this approach is necessary to quantify the accuracy of the
student dropout methodology which is being developed
1: Start
through Fuzzy ANN and decision tree algorithms. The
2: Set min=0.5, max=0.5 measurement of the accuracy of the approach will be useful
3: for i=0 to size of SLST in determining if these algorithms have been accurately
implemented which will be evident through the scores
4: TMPLST = SLST[i] [ TMPLST = Temporary Set] achieved by the precision and recall metrics.
5: PRSCRE= TMPLST [1] [ Score]
Performance Evaluation based on Precision and Recall
6: IF (PRSCRE < min) The performance metrics through precision and recall have
7: min= PRSCRE been assessed through the use of intensive experimentation
that has been performed on the proposed methodology.
8: IF(PRSCRE >max)
These performance parameters are useful in determining the
9: max= PRSCRE real accuracy of the evaluation and the actual performance of
10: end for the methodology. The student dropout detection is highly
useful and can only be viable with a respectable accuracy of
11: RANGE1=0 , RANGE2=0 ,RLIST= ∅ [Rule List]
the identification through the methodology prescribed in
12: DF=( max-min)/5 [ DF= Diffrence Distance ] this research work.
13: for i=0 to 5 The in-depth information regarding the performance of the
14: RANGE1=min approach has been achieved through the precision and recall
15: RANGE2=RANGE1+DF parameters where each of these parameters is useful in
identifying a different aspect of performance of the
16: min=RANGE2 methodology. The precision parameter is useful for
17: TLST[0]= RANGE1 [TLST= Temporary List] determining the relative performance of the student dropout
18: TLST[1]= RANGE2 prediction on the other hand recall extracts the absolute
accuracy for the student dropout prediction.
19: FCLST = FCLST + TLST
The precision for this methodology has been conceived as
20: end for
the division of the number of accurate student dropout
21: return FCLST predictions by the total number of expected student dropout
22: Stop predictions. The recall parameter is on the other hand the
absolute accuracy which is derived by the ratio of total
_________________________________ number of student dropout prediction versus the total
The neurons for the artificial neural network are then number of expected student dropout prediction.
classified according to the given rules above depending on This has been mathematically conveyed through the
the classified certified set with labels provided such as very equations 4 and 5 given below.
high, high, medium, low and very low. These neurons are
then subjected to affective evaluation of the hidden layer and A = The number of correctly predicted student dropout
the resultant probability achieved effectively determines the B= The number of incorrectly predicted student dropout
dropout rate for the desired level according to the user
C = The number of student dropout not predicted
Step 4: Decision Tree – The probability values achieved from
the previous step of fuzzy artificial neural networks is taken So, precision can be defined as
as an input to this step of the methodology. The Decision tree Precision = (A / (A+ B)) *100
approach is one of the most powerful and highly accurate
classification algorithms that completely categorize the data Recall = (A / (A+ C)) *100
and segregate is effectively. The decision tree approach The in-depth experimentation done on the prescribed
utilizes the if-then rules to accomplish its goals of approach for the measurement of precision and recall
classification based on the threshold values applied for parameters has been listed in the table 1 below.
execution. The probability scores related to the student
dropout scenario have been subjected to the classification to
achieve highly accurate classification of the dropout
predictions. This achieves the student dropout results that
are displayed to the user through the interactive user
interface.
Table 1: Precision and Recall Measurement Table for
RESULTS AND DISCUSSIONS the performance of student dropout prediction
The proposed system for student dropout prediction
through the use of Fuzzy Artificial Neural Networks has been

@ IJTSRD | Unique Paper ID – IJTSRD43700 | Volume – 5 | Issue – 4 | May-June 2021 Page 1552
 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
[3] J. Figueroa-Cañas and T. Sancho-Vinuesa, "Early
Prediction of Dropout and Final Exam Performance in
an Online Statistics Course," in IEEE Revista
Iberoamericana de Tecnologias del Aprendizaje, vol.
15, no. 2, pp. 86-94, May 2020, doi:
10.1109/RITA.2020.2987727.
[4] M. N. Mustafa, L. Chowdhury and M. S. Kamal,
"Students dropout prediction for intelligent system
from tertiary level in developing country," 2012
International Conference on Informatics, Electronics
& Vision (ICIEV), Dhaka, 2012, pp. 113-118, doi:
10.1109/ICIEV.2012.6317441.
[5] L. Wang and H. Wang, "Learning Behavior Analysis
and Dropout Rate Prediction Based on MOOCs Data,"
Figure 2: Comparison of Precision and Recall for the
2019 10th International Conference on Information
performance of student dropout prediction
Technology in Medicine and Education (ITME),
The outcomes achieved in the table are effectively used to Qingdao, China, 2019, pp. 419-423, doi:
draw a plot graph for precision and recall parameters for 10.1109/ITME.2019.00100.
graphical representation in the figure 2 above. As it is [6] N. Shiratori, "Derivation of Student Patterns in a
evident from the tabulated values and the outcome of the Preliminary Dropout State and Identification of
line graph, that the precision and recall parameters achieved Measures for Reducing Student Dropouts," 2018 7th
for this methodology are well within respectable limits. The International Congress on Advanced Applied
value of precision and recall as 84.06 and 83.54 are Informatics (IIAI-AAI), Yonago, Japan, 2018, pp. 497-
indicative of an accurate implementation of Fuzzy ANN and 500, doi: 10.1109/IIAI-AAI.2018.00108.
Decision Tree approaches that have been utilized for [7] D. Ktoridou and E. Epaminonda, "Measuring the
accurate student dropout prediction in the proposed compatibility between engineering students'
methodology. personality types and major of study: A first step
CONCLUSION towards preventing engineering education dropouts,"
This research article has been effective in the analysis of 2014 IEEE Global Engineering Education Conference
predominant researchers on the topic of dropout prediction. (EDUCON), Istanbul, 2014, pp. 192-195, doi:
The dropouts are highly undesirable occurrence that usually 10.1109/EDUCON.2014.6826089.
is a loss for the student as well as the university involved. [8] R. Timaran Pereira and J. Caicedo Zambrano,
The dropouts can hurt the universities negatively and could "Application of Decision Trees for Detection of
be one of the reasons for the increased dissatisfaction of the Student Dropout Profiles," 2017 16th IEEE
students with the course. These problems can only be solved International Conference on Machine Learning and
through to the effective implementation of a dropout Applications (ICMLA), Cancun, 2017, pp. 528-531, doi:
prediction approach. Therefore this research article has been 10.1109/ICMLA.2017.0-107.
effective in achieving methodology utilizing machine [9] A. Ortigosa, R. M. Carro, J. Bravo-Agapito, D. Lizcano, J.
learning approaches for the purpose of dropout prediction. J. Alcolea and Ó. Blanco, "From Lab to Production:
The input dataset is first effectively preprocessed and Lessons Learnt and Real-Life Challenges of an Early
labeled before providing it to the K Nearest neighbor for the Student-Dropout Prevention System," in IEEE
clustering. The KNN approach clusters the data using the Transactions on Learning Technologies, vol. 12, no. 2,
certified labels. The generated clusters are provided to the pp. 264-277, 1 April-June 2019, doi:
Fuzzy Artificial Neural Networks as an input which 10.1109/TLT.2019.2911608.
processes the clusters using the fuzzy crisp values to achieve [10] K. Murakami et al., "Predicting the Probability of
the predictions. These predictions are then effectively Student Dropout through EMIR Using Data from
classified using the if-then rules of the Decision tree Current and Graduate Students," 2018 7th
approach. The outcomes indicate the effective International Congress on Advanced Applied
implementation of the student dropout methodology. Informatics (IIAI-AAI), Yonago, Japan, 2018, pp. 478-
The future research directions can be focused on the 481, doi: 10.1109/IIAI-AAI.2018.00103.
realization of the student dropout methodology to be [11] K. J. de O. Santos, A. G. Menezes, A. B. de Carvalho and
implemented in a web application for easier access to the C. A. E. Montesco, "Supervised Learning in the Context
required authorities. of Educational Data Mining to Avoid University
Students Dropout," 2019 IEEE 19th International
References
Conference on Advanced Learning Technologies
[1] T. Hasbun, A. Araya and J. Villalon, "Extracurricular
(ICALT), Macei??, Brazil, 2019, pp. 207-208, doi:
Activities as Dropout Prediction Factors in Higher
10.1109/ICALT.2019.00068.
Education Using Decision Trees," 2016 IEEE 16th
[12] R. Lottering, R. Hans and M. Lall, "A model for the
International Conference on Advanced Learning
identification of students at risk of dropout at a
Technologies (ICALT), Austin, TX, 2016, pp. 242-244,
university of technology," 2020 International
doi: 10.1109/ICALT.2016.66.
Conference on Artificial Intelligence, Big Data,
[2] D. Sun, Y. Mao, J. Du, P. Xu, Q. Zheng and H. Sun, "Deep
Computing and Data Communication Systems
Learning for Dropout Prediction in MOOCs," 2019
(icABCD), Durban, South Africa, 2020, pp. 1-8, doi:
Eighth International Conference on Educational
10.1109/icABCD49160.2020.9183874.
Innovation through Technology (EITT), Biloxi, MS,
USA, 2019, pp. 87-90, doi: 10.1109/EITT.2019.00025.

@ IJTSRD | Unique Paper ID – IJTSRD43700 | Volume – 5 | Issue – 4 | May-June 2021 Page 1553