Predicting Student Performance Based On Clustering And Classification

Abstract

In today's world the education field is growing, developing widely and becoming one of the most crucial industries. The data available in the educational field can be studied using educational data mining so that the unseen knowledge can be obtained from it. In this paper, various data mining approaches like Clustering, classification and regression our used to predict the students' performance in examination in advance, so that necessary measures can be taken to improvise on their performance to score better marks. A hybrid approach of Enhanced K-strange points clustering algorithm and Naïve Bayes classification algorithm is presented implemented and compared it with existing hybrid approach which is K-means clustering algorithm and Decision tree. Finally, to predict student performance, multiple linear regression is used. The results obtained after the implementation may be useful for instructor as well as students. This work will help in taking appropriate decision to improve student's performance. 1. INTRODUCTION Students tend to drop out or have a significant decrease in its academic performance. By predicting student performance, instructors can help to improve student performance in the examination and significantly reduce drop out ratio from college, which will enhance the performance of college. In paper [1], K-means clustering algorithm and Decision tree has been used to predict student performance. But K-means clustering algorithm has a limitation that in case clusters or centroid does not converge that it can go into infinite iteration hence in this work, Enhanced K-strange point clustering algorithm is used since iteration depends upon number of clusters. The disadvantage of Decision Tree classification algorithm is that it is not considering all the attributes of the dataset which is essential to predict student performance hence in this paper, Naïve Bayes classification algorithm is proposed as this algorithm considers all the attributes while computing the result. 2. LITERATURE SURVEY In paper[1],K-means clustering algorithm is used to form the clusters. The algorithm was applied on the student training data set, then three clusters were formed namely "High", "Medium" and "Low", according to their new grade. The new grade is calculated from the previous semester grade that means external assessment and internal assessment. Then Decision tree was applied to make correct decisions about the student's performance, which can use by the instructor to take the necessary steps. In paper [2], the student performance is prediction is carried out using K-means clustering algorithms and decision trees, the results and analysis was done in WEKA tool. K-means algorithm was applied on the same dataset using WEKA tool. The decision tree algorithm was used to do the prediction which was displayed in tree-like structure. 143 students were classified as passed and 30 as failed which was true as per the original dataset. K-means clustering algorithm is used on the student's data and then students have been clustered based on their class performance, sessionals and attendance in class [3]. Centroids are calculated from the educational data set taking K-clusters. This study helps in identifying students who are short of attendance and have shown poor performance in sessionals. Paper [4] provides an enhancement to K Strange points clustering algorithm by correcting the location of the third strange point by trying to place it almost maximally and equally spaced both from Kmin and Kmax.This results in more accurate clusters.