IRJET-Student Performance Prediction for Education Loan System
0 likes•69 views
This document presents a student performance prediction system that uses machine learning algorithms to predict how well students will perform in their degree programs based on their current and past academic performance data. The system uses a bi-layered architecture with a base predictor layer and an ensemble predictor layer. The base predictor layer makes local predictions about student performance using various predictors trained on student data features. The ensemble layer synthesizes these local predictions along with previous overall predictions to make a final performance prediction. Latent factor models are used to identify relevant course subjects. The system aims to help banking systems assess student loan eligibility by predicting their likelihood of satisfactory and timely degree completion.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2388
Student Performance Prediction for Education Loan System
Radhika Kale
Vidya Pratishthan’s Kamalnayan Bajaj Institute of Engineering and Technology, Baramati
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In education system evaluation and prediction of
student performance is a challenging task. Usually, a degree
program consist of term and course architecturewhereaterm
is separated in semesters and courses have set of subjects on
basis of background of student and interest. Recently, student
background and performance are used toprovidethefinancial
loan by the banking system for the completion of degree
program. Here, banking system serviceaseducationloanrelies
on student academic performance in a term. The student
performance prediction model is based on base predictor and
ensemble predictor, including data set. A data-driven
approach based on latent factor models and probabilistic
matrix factorization is used to discover course relevance,
which is important for constructing efficient base predictors.
This way banking system gets the assurance of successful
degree program. Current performanceisalsoimportantwhich
add the visualization of future student performance. Very few
searches have been made for predictingstudents'performance
by completion of degrees, as most searches focus on
performance prediction on historical log dataset. Machine
learning approaches are used for automatic prediction of
students' performance in degree program.
Key Words: Course Relevance, Data driven approach,
Latent factor, Prediction of student performance,
personalized education, Probabilistic Matrix
Factorization
1. INTRODUCTION
Data mining is a major thrust area to predictthefuturescope
in research domain, including education, business, industry,
forensic science, health care, cyber security,etc.Ineducation
system it is necessary to build a system that can keep track
of students' performance which accurately, predicts
students' future performance. Existing approach shows
difficulties to cover the diversities of student's educational
background and it becomes critical for undefined set of
course. In case of relevant course student background
information is considered which helps to estimate the
relevant course in a term. Traditional existing system,
Intelligent Tutoring system and Massiveopenonlinecourses
are working on past records. It is difficult to predict the
student performance over on-going records so, a prediction
model required. Here, it usesmachine learning approachfor
prediction of performance which speedsuptheperformance
and also reduces the computation time. We develop a novel
algorithm for making predictions based on students’
progressive performance states. This type of model consists
of two layers of predictor which is named as base predictor
and ensemble predictor for performance evaluation. In the
base layer, multiple base predictors make local predictions
given the snapshot of the student’s current performance
state in each academic term. In the ensemble layer, an
ensemble predictor issues a prediction of the student’s
future performance by synthesizing the local predictions
results as well as the previous-term ensemble prediction.
The cascading of ensemble predictor over academic terms
enablesthe incorporationof students’evolvingprogressinto
the prediction while keeping the complexity low. This
system also form cluster of courses for finding relevant
courses to train the data according to student and the
information that a student provides.
2. LITERATURE SURVEY
Rahel Bekele and Wolfgang Menzelproposedtheimportance
of accurate estimation of student’s future performance is
essential in order to provide the student with adequate
assistance in the learning process. To this end, this research
aimed at investigating the use of Bayesian networks for
predicting performance of a student, based on values of
some identified attributes. This present empirical
experiment on the prediction of performance with a dataset
of high school students containing 8 attributes. The paper
demonstratesan application of the Bayesian approachinthe
field of education and shows that the Bayesian network
classifier hasa potential to be used as a tool for predictionof
student performance [2]. C.MARQUEZ-VERA proposed to
apply data mining techniques to predict school failure.
Several experiments have been carried out in an attempt to
improve accuracy in the prediction of final student
performance and specifically of which students might fail.
The outcomes of each one of these approaches using the 10
classification algorithms and 10-fold cross validation is
shown and compared in order to select the best approach to
this problem. This research is failed to develop our own
classification algorithm using grammar-based genetic
programming and cost sensitive classification for
comparison versus other classification algorithms [3]. Hao
Cen, Kenneth Koedinger, and Brian Junker proposes a semi-
automated method for improving a cognitive model called
Learning Factors Analysis that combines a statistical model,
human expertise and a combinatorial search. This methodis
used to evaluate an existing cognitive model and to generate
and evaluate alternative models. To use the method for
datasetsfrom other tutorsto discover itspotentialformodel
and tutor improvement [4]. Nguyen Thai-Nghe, Tomas
Horvath and Lars Schmidt-Thieme proposes to take into
account the sequential effect, this work proposes a novel
approach which uses tensor factorization for forecasting
student performance. With this approach, we can
personalize the prediction for each student given the task;
thus, it can also be used for recommending the tasks to the
students. Experimental results on two large data sets show](https://image.slidesharecdn.com/irjet-v5i6446-180813070004/85/IRJET-Student-Performance-Prediction-for-Education-Loan-System-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2389
that incorporating forecasting techniques into the
factorization process is a promising approach [5]. Mingyu
Feng, Neil Heffernan, Kenneth Koedinger proposed the
assessing student math proficiency is to use data that our
system collects through its interactions with students to
estimate their performance on an end-of-year high stakes
state test. This result show that we can do a reliably better
job predicting student end-of-year exam scores by
leveraging the interaction data, and the model based ononly
the interaction. Continues assessment systems can perform
better as compared to the system proposed in this paper[6].
Man-Ching Yuen, Irwin King, Kwong Sak Leung proposed a
Task Recommendation (TaskRec) framework based on a
united probabilistic matrix factorization, aiming to
recommend tasks to workers in dynamic scenarios. Unlike
traditional recommendation systems, workers do not
provide their ratings on tasks in crowd sourcing systems,
thusthey infer user ratingsfrom their interacting behaviors
[7].
3. SYSTEM ARCHITECTURE
Fig: Block diagram of system
Fig shows the details of bi-layered architecture where
system takes student's details at the time of course
registration. There is a test data of student historical record
i.e. their performance reports and logs. The feature
extraction process is done on student's provided data. This
data is compared with all students’ data from test setoftheir
courses.
In system, student provides the information of his/her
educational background. This information along with the
reports is used to predict results using two different layers
of predictors. Those are as given below:
1. Base predictor layer:
This layer takes the information of students and performs
feature extraction. This layer consists of number of
predictors which predicts performance individually. For
each base predicator denote the predictionresultforstudent
given the student’s static feature and the current
performance state. The base predictors are trained using a
dataset consisting of all student data. Learning the base
predictor is done offline.
2. Ensemble layer:
This layer does the noise reduction part and finds relevant
coursesfor student'sperformance prediction.Thepredicted
output is then again passed to the base predictors for
adaptation and consideration of student'sperformancestate
in future prediction. The ensemble predictor synthesizesthe
previousensemble output and outputof the base predictors
and makes a final prediction.
The ensemble predictor is learned using student data.
Learning the ensemble predictorsis done online.Theoutput
predicted by ensemble layer is further passed to the base
predictor layer. This ensures that the current performance
state of student is taken into consideration for next
prediction.
The naïve bayes classifier is used for classification and
prediction purpose. The latent factor model is used for
suggest subject which are not covered by previousdegreeor
year.
The probabilistic matrix factorization is used to learn the
course correlation which is performed on the student
dataset. This model generates a report of performance
prediction. This report helps to bank to ensure student’s
satisfactory and on-time graduation.](https://image.slidesharecdn.com/irjet-v5i6446-180813070004/85/IRJET-Student-Performance-Prediction-for-Education-Loan-System-2-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2390
4. ALGORITHM
Algorithm 1: [1]
Descriptions of symbol that are used in algorithm are
following:
L(h) : Loss of base predictor ,
gi : Group of i student
ht : Base predictor value ,
ⱺi : Static function
Zth,i: Prediction value of grade of targeted course,
ft : Ensemble prediction,
wi(ht) : Weight vector of base predictor,
G : SAT scores/Background,
H : Number of base predictor,
Ln(h/g) : Cumulative loss,
l(y’ ; y) : Loss function,
Nk(j) : Targeted course of cluster
Algorithm 2: Latent factor
1. Input: Subject List (S), Course Name (c)
2. Output: Latent Factor for Student
3. Latent Factor ← Φ
4. for each Course c ∈ C
5. for each Subject s ∈ c
6. subjects subs ← findRelevantSubjects(s)
7. for each Subject sub ∈ subs
8. if(!(sub ∈ S))
9. Latent Factor ← sub
10. end if
11. end for each
12. end for each
13. end for each
14. return Latent Factor
In this system training and testing dataset are used to
trained system. The training dataset is of different courses
along with their prerequisite courses and subjects. Here in
latent factor,find out the relevant subjects for the specific
courses and that is the output of the latent factor.
Here in this algorithm, first trained the system using arff, we
load the arrff, then we pass the arrff to the classifier to
predict the output.
In the above algorithmon line 1 givessubjects and course as
a input then we find the relevant subjects relevant to the
selected course.
Finally this system is a student performance prediction
system on that we can predict student performance for
future use.](https://image.slidesharecdn.com/irjet-v5i6446-180813070004/85/IRJET-Student-Performance-Prediction-for-Education-Loan-System-3-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2392
7. CONCLUSIONS
A novel method for predicting student'sfuture performance
in degree programs given their current and past
performance. This performance is also affected on banking
services to get loan for student’seducation. Ensemble-based
progressive prediction architecture is developing to
incorporate student's evolving performance into the
prediction.
REFERENCES
[1] Jie Xu, Member, IEEE, Kyeong Ho Moon and Mihaela
vander Schaar "A Machine Learning Approach for Tracking
and Predicting Student Performance in Degree Programs"
DOI 10.1109/JSTSP.2017.2692560, IEEE Journal.
[2] R. Bekele and W. Menzel, "A Bayesian approachtopredict
performance of a student (bapps): A case with Ethiopian
students”, algorithms, vol. 22, no. 23, p. 24, 2005.
[3] C. Marquez-Vera, C. Romero, and S. Ventura, “Predicting
school failure using data mining,” in EducationalDataMining
2011, 2010.
[4] H. Cen, K. Koedinger, and B. Junker, “Learning factors
analysis–a general method for cognitive model evaluation
and improvement,” in International Conference on
Intelligent Tutoring Systems. Springer, 2006, pp. 164–175.
[5] N. Thai-Nghe, T. Horv´ath, and L. Schmidt Thieme,
“Factorization modelsfor forecasting studentperformance,”
in Educational Data Mining 2011, 2010.
[6] M. Feng, N. Heffernan, and K. Koedinger, “Addressing the
assessment challenge with an online system that tutors as it
assesses,” User Modeling and User-Adapted Interaction, vol.
19, no. 3, pp. 243–266, 2009.
[7] M.-C. Yuen, I. King, and K.-S. Leung,"Task
recommendation in crowd sourcingsystems,”inProceedings
of the First International Workshop on Crowd sourcing and
Data Mining. ACM, 2012, pp. 22–26.
[8] C. G. Brinton and M. Chiang, “Mooc performance
prediction via click stream data and social learning
networks,” in 2015 IEEE Conference on Computer
Communications (INFOCOM). IEEE, 2015, pp. 2299– 2307.
[9] KDD Cup, “Educational data minding challenge,”
https://pslcdatashop:web:cmu:edu/KDDCup/, 2010.13
[10] C. Marquez-Vera, C. Romero, and S. Ventura, “Predicting
school failure using data mining,” in EducationalDataMining
2011, 2010.](https://image.slidesharecdn.com/irjet-v5i6446-180813070004/85/IRJET-Student-Performance-Prediction-for-Education-Loan-System-5-320.jpg)
IRJET-Student Performance Prediction for Education Loan System
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072 © 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2388 Student Performance Prediction for Education Loan System Radhika Kale Vidya Pratishthan’s Kamalnayan Bajaj Institute of Engineering and Technology, Baramati ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - In education system evaluation and prediction of student performance is a challenging task. Usually, a degree program consist of term and course architecturewhereaterm is separated in semesters and courses have set of subjects on basis of background of student and interest. Recently, student background and performance are used toprovidethefinancial loan by the banking system for the completion of degree program. Here, banking system serviceaseducationloanrelies on student academic performance in a term. The student performance prediction model is based on base predictor and ensemble predictor, including data set. A data-driven approach based on latent factor models and probabilistic matrix factorization is used to discover course relevance, which is important for constructing efficient base predictors. This way banking system gets the assurance of successful degree program. Current performanceisalsoimportantwhich add the visualization of future student performance. Very few searches have been made for predictingstudents'performance by completion of degrees, as most searches focus on performance prediction on historical log dataset. Machine learning approaches are used for automatic prediction of students' performance in degree program. Key Words: Course Relevance, Data driven approach, Latent factor, Prediction of student performance, personalized education, Probabilistic Matrix Factorization 1. INTRODUCTION Data mining is a major thrust area to predictthefuturescope in research domain, including education, business, industry, forensic science, health care, cyber security,etc.Ineducation system it is necessary to build a system that can keep track of students' performance which accurately, predicts students' future performance. Existing approach shows difficulties to cover the diversities of student's educational background and it becomes critical for undefined set of course. In case of relevant course student background information is considered which helps to estimate the relevant course in a term. Traditional existing system, Intelligent Tutoring system and Massiveopenonlinecourses are working on past records. It is difficult to predict the student performance over on-going records so, a prediction model required. Here, it usesmachine learning approachfor prediction of performance which speedsuptheperformance and also reduces the computation time. We develop a novel algorithm for making predictions based on students’ progressive performance states. This type of model consists of two layers of predictor which is named as base predictor and ensemble predictor for performance evaluation. In the base layer, multiple base predictors make local predictions given the snapshot of the student’s current performance state in each academic term. In the ensemble layer, an ensemble predictor issues a prediction of the student’s future performance by synthesizing the local predictions results as well as the previous-term ensemble prediction. The cascading of ensemble predictor over academic terms enablesthe incorporationof students’evolvingprogressinto the prediction while keeping the complexity low. This system also form cluster of courses for finding relevant courses to train the data according to student and the information that a student provides. 2. LITERATURE SURVEY Rahel Bekele and Wolfgang Menzelproposedtheimportance of accurate estimation of student’s future performance is essential in order to provide the student with adequate assistance in the learning process. To this end, this research aimed at investigating the use of Bayesian networks for predicting performance of a student, based on values of some identified attributes. This present empirical experiment on the prediction of performance with a dataset of high school students containing 8 attributes. The paper demonstratesan application of the Bayesian approachinthe field of education and shows that the Bayesian network classifier hasa potential to be used as a tool for predictionof student performance [2]. C.MARQUEZ-VERA proposed to apply data mining techniques to predict school failure. Several experiments have been carried out in an attempt to improve accuracy in the prediction of final student performance and specifically of which students might fail. The outcomes of each one of these approaches using the 10 classification algorithms and 10-fold cross validation is shown and compared in order to select the best approach to this problem. This research is failed to develop our own classification algorithm using grammar-based genetic programming and cost sensitive classification for comparison versus other classification algorithms [3]. Hao Cen, Kenneth Koedinger, and Brian Junker proposes a semi- automated method for improving a cognitive model called Learning Factors Analysis that combines a statistical model, human expertise and a combinatorial search. This methodis used to evaluate an existing cognitive model and to generate and evaluate alternative models. To use the method for datasetsfrom other tutorsto discover itspotentialformodel and tutor improvement [4]. Nguyen Thai-Nghe, Tomas Horvath and Lars Schmidt-Thieme proposes to take into account the sequential effect, this work proposes a novel approach which uses tensor factorization for forecasting student performance. With this approach, we can personalize the prediction for each student given the task; thus, it can also be used for recommending the tasks to the students. Experimental results on two large data sets show
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072 © 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2389 that incorporating forecasting techniques into the factorization process is a promising approach [5]. Mingyu Feng, Neil Heffernan, Kenneth Koedinger proposed the assessing student math proficiency is to use data that our system collects through its interactions with students to estimate their performance on an end-of-year high stakes state test. This result show that we can do a reliably better job predicting student end-of-year exam scores by leveraging the interaction data, and the model based ononly the interaction. Continues assessment systems can perform better as compared to the system proposed in this paper[6]. Man-Ching Yuen, Irwin King, Kwong Sak Leung proposed a Task Recommendation (TaskRec) framework based on a united probabilistic matrix factorization, aiming to recommend tasks to workers in dynamic scenarios. Unlike traditional recommendation systems, workers do not provide their ratings on tasks in crowd sourcing systems, thusthey infer user ratingsfrom their interacting behaviors [7]. 3. SYSTEM ARCHITECTURE Fig: Block diagram of system Fig shows the details of bi-layered architecture where system takes student's details at the time of course registration. There is a test data of student historical record i.e. their performance reports and logs. The feature extraction process is done on student's provided data. This data is compared with all students’ data from test setoftheir courses. In system, student provides the information of his/her educational background. This information along with the reports is used to predict results using two different layers of predictors. Those are as given below: 1. Base predictor layer: This layer takes the information of students and performs feature extraction. This layer consists of number of predictors which predicts performance individually. For each base predicator denote the predictionresultforstudent given the student’s static feature and the current performance state. The base predictors are trained using a dataset consisting of all student data. Learning the base predictor is done offline. 2. Ensemble layer: This layer does the noise reduction part and finds relevant coursesfor student'sperformance prediction.Thepredicted output is then again passed to the base predictors for adaptation and consideration of student'sperformancestate in future prediction. The ensemble predictor synthesizesthe previousensemble output and outputof the base predictors and makes a final prediction. The ensemble predictor is learned using student data. Learning the ensemble predictorsis done online.Theoutput predicted by ensemble layer is further passed to the base predictor layer. This ensures that the current performance state of student is taken into consideration for next prediction. The naïve bayes classifier is used for classification and prediction purpose. The latent factor model is used for suggest subject which are not covered by previousdegreeor year. The probabilistic matrix factorization is used to learn the course correlation which is performed on the student dataset. This model generates a report of performance prediction. This report helps to bank to ensure student’s satisfactory and on-time graduation.
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072 © 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2390 4. ALGORITHM Algorithm 1: [1] Descriptions of symbol that are used in algorithm are following: L(h) : Loss of base predictor , gi : Group of i student ht : Base predictor value , ⱺi : Static function Zth,i: Prediction value of grade of targeted course, ft : Ensemble prediction, wi(ht) : Weight vector of base predictor, G : SAT scores/Background, H : Number of base predictor, Ln(h/g) : Cumulative loss, l(y’ ; y) : Loss function, Nk(j) : Targeted course of cluster Algorithm 2: Latent factor 1. Input: Subject List (S), Course Name (c) 2. Output: Latent Factor for Student 3. Latent Factor ← Φ 4. for each Course c ∈ C 5. for each Subject s ∈ c 6. subjects subs ← findRelevantSubjects(s) 7. for each Subject sub ∈ subs 8. if(!(sub ∈ S)) 9. Latent Factor ← sub 10. end if 11. end for each 12. end for each 13. end for each 14. return Latent Factor In this system training and testing dataset are used to trained system. The training dataset is of different courses along with their prerequisite courses and subjects. Here in latent factor,find out the relevant subjects for the specific courses and that is the output of the latent factor. Here in this algorithm, first trained the system using arff, we load the arrff, then we pass the arrff to the classifier to predict the output. In the above algorithmon line 1 givessubjects and course as a input then we find the relevant subjects relevant to the selected course. Finally this system is a student performance prediction system on that we can predict student performance for future use.
- 4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072 © 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2391 5. FLOW CHART 6. RESULTS Fig: Prediction result
- 5. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072 © 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2392 7. CONCLUSIONS A novel method for predicting student'sfuture performance in degree programs given their current and past performance. This performance is also affected on banking services to get loan for student’seducation. Ensemble-based progressive prediction architecture is developing to incorporate student's evolving performance into the prediction. REFERENCES [1] Jie Xu, Member, IEEE, Kyeong Ho Moon and Mihaela vander Schaar "A Machine Learning Approach for Tracking and Predicting Student Performance in Degree Programs" DOI 10.1109/JSTSP.2017.2692560, IEEE Journal. [2] R. Bekele and W. Menzel, "A Bayesian approachtopredict performance of a student (bapps): A case with Ethiopian students”, algorithms, vol. 22, no. 23, p. 24, 2005. [3] C. Marquez-Vera, C. Romero, and S. Ventura, “Predicting school failure using data mining,” in EducationalDataMining 2011, 2010. [4] H. Cen, K. Koedinger, and B. Junker, “Learning factors analysis–a general method for cognitive model evaluation and improvement,” in International Conference on Intelligent Tutoring Systems. Springer, 2006, pp. 164–175. [5] N. Thai-Nghe, T. Horv´ath, and L. Schmidt Thieme, “Factorization modelsfor forecasting studentperformance,” in Educational Data Mining 2011, 2010. [6] M. Feng, N. Heffernan, and K. Koedinger, “Addressing the assessment challenge with an online system that tutors as it assesses,” User Modeling and User-Adapted Interaction, vol. 19, no. 3, pp. 243–266, 2009. [7] M.-C. Yuen, I. King, and K.-S. Leung,"Task recommendation in crowd sourcingsystems,”inProceedings of the First International Workshop on Crowd sourcing and Data Mining. ACM, 2012, pp. 22–26. [8] C. G. Brinton and M. Chiang, “Mooc performance prediction via click stream data and social learning networks,” in 2015 IEEE Conference on Computer Communications (INFOCOM). IEEE, 2015, pp. 2299– 2307. [9] KDD Cup, “Educational data minding challenge,” https://pslcdatashop:web:cmu:edu/KDDCup/, 2010.13 [10] C. Marquez-Vera, C. Romero, and S. Ventura, “Predicting school failure using data mining,” in EducationalDataMining 2011, 2010.