Prediction of Academic Performance Using Gravitational Search Based Neural Network Algorithm

Chaudhary, Somendra and Imran, Abdul and Kolekar, Sucheta (2017) Prediction of Academic Performance Using Gravitational Search Based Neural Network Algorithm. In: International Conference on Inventive Computing and Informatics, 23/11/2017, Coimbatur, Tamilnadu.

[img] PDF
941.pdf - Published Version
Restricted to Registered users only

Download (164kB) | Request a copy

Abstract

Education is the way towards encouraging learning, or the procurement of information, abilities, values, convictions, & propensities. Education often happens under the direction of instructors. Academic performance is the result of education - the degree to which a pupil, educator or institute has accomplished their educative targets. Education is a critical issue with respect to the advancement of a nation, particularly in India, where tutoring is a component firmly connected with social mobility; accordingly, it is of extraordinary enthusiasm to recognize the pupils who are at hazard of failing, at the earliest opportunity, & also to understand which variables affect this. A Data Mining model is an appropriate instrument to include these assignments. An approach where supervised learning, learning where a training set of correctly identified observations is available is ideal. Statistical classification or classification, which is an instance of supervised learning is chosen as the method to achieve aforementioned tasks. A data set from two Portuguese schools, a country where educational & social mobility are closely related, much like India, was chosen. Broadly, three classification algorithms are used to construct data mining models: Na¨ıve Bayes Classifier, a Decision Tree Classifier (C4.5) & Neural Networks (Feedforward Neural Networks) using Gravitational Search Algorithm. Firstly, a C4.5 Decision Tree in Java is used to test & train the data for both pruned & unpruned trees. Further, both Na¨ıve Bayes Classifier & C4.5 are used to train models using cross validation & cost matrices. Lastly, a Feedforward Neural Network is trained for the dataset in which weights are updated using GSA which further takes in consideration, the error of FNN. Neural Networks (Feedforward Neural Networks) using Gravitational Search Algorithm tends to have better accuracy than its counterparts.

Item Type: Conference or Workshop Item (Paper)
Uncontrolled Keywords: Educational Data Mining (EDM), Classification, Dropout, Prediction, Attrition, Optimization, Heuristic Search Algorithms, Gravitational Search Algorithm, FNN, Neural network, Learning Neural Network, Feedforward Neural Networks, Law of gravity
Subjects: Engineering > MIT Manipal > Information and Communication Technology
Depositing User: MIT Library
Date Deposited: 06 Dec 2017 09:28
Last Modified: 06 Dec 2017 09:28
URI: http://eprints.manipal.edu/id/eprint/150149

Actions (login required)

View Item View Item