Predicting Academic Performance of Engineering Students Using Ensemble Method

Abstract

 One of the common problems that most of the engineering institutions face in recent times is poor academic results. The statistics of the IOE semester result show that since 2009 A.D., the average pass percentage has been reducing in an average from 50% to 40% and moving towards decreasing scenarios. The study aims to predict an engineering student's academic performance based on their past educational records, demographic factors, family backgrounds, and other related factors. Firstly, a predictive model is built using the traditional classifiers Decision Tree, SVM, and Linear Regression, which had shown good performance in similar types of study. After that, we have implemented one of the popular ensemble Methods, voting, which is known for improving the individual classifier's performance. Voting classifier combines the predictions of base classifiers by averaging those predictions. The result revealed that the accuracy, precision, recall, and F1 score had been considerably improved by using ensemble voting than that of the individual classifiers. The data used in the study was collected directly from the hard copy personal files of each pass out student of Paschimanchal Engineering Campus, Pokhara between the years 2004 to 2015 AD.