Predicting Students’ Performance and Its Influential Factors Using Ensemble Approach - LRN

Abstract

The effectiveness of any educational institution can be measured by the academic growth of their students. In light of emerging technologies such as artificial intelligence, machine learning and data mining, institutions are integrating technology into conventional teaching approaches. In today’s education system, prediction of students’ performance is extremely important. Predicting student performance in advance can help the students, teachers as well as the institutions. Prediction of students’ performance has been crucial issue to forecast whether a student will complete his/her study within stipulated time period or not.Students should be advised well in advance to concentrate their efforts in specific area in order to improve their academic achievement. The objective of this study is to identify uncover/hidden factors that could assist in predicting the students’ performance. We used educational data mining (EDM), which is the collection of techniques used to uncover hidden patterns in massive amount of existing data. In this study, primary data set has been collected from the students of different schools of Jalandhar city . The size of dataset is 399 records with 17 attributes. Features that are redundant or unnecessary are removed from the dataset via feature selection approaches, which have been used to improve the prediction accuracy of standalone classifiers while lowering overall costs. We used and analyzed the different machine learning algorithms such as SVM, Logistic Regression, Naïve Bayes (NB), Random Forest, Boosting, Bagging and Stacking. To lessen sample related bias in our investigation, we used 5-fold cross validation. One of the causes of accurate prediction outcomes is this. The whole dataset is divided into two sets of training set and testing set 75:25 proportions respectively. In this research, an ensemble model LRN proposed to identify prediction of students’ performance. For better performance, with hyper-parameter tuning three base classifiers are gathered and added to the  proposed ensemble model LRN (Logistic Regression+Random Forest+Naive Bayes). By using ensemble techniques; we will have a good result that demonstrates the dependability of the proposed model. Our proposed model gave accuracy 96% which is the highest among SVM, Logistic Regression, Naive Bayes, Random Forest and other boosting algorithms.