Using Traditional Machine Learning Algorithms and SMOTE Technique to Estimate Student’s Academic Performance in Higher Education

Abstract

Predictive analysis applications have become a revolutionary topic in higher education in the following days. Predictive analysis will use analytical models, which include machine learning applications, to help in producing the greater quality performance and useful student data at different stages of education. Most people know that a student's grade is the most important performance indicator that teachers can use to track their academic progress. Many learning algorithms have been proposed in the education field over the past decade. However, dealing with unequal databases in order to improve the efficiency of predicting student marks causes significant difficulties. Random Forest (RF), Naive Bayes (NB), Decision Tree (J48), K-Nearest Neighbour (kNN), Logistic Regression (LR), Support vector Machine (SVM), and a hybrid model that combines Random Forest and the XGBoost algorithm are all evaluated for accuracy. By using feature selection and Synthetic Minority Oversampling Technique (SMOTE) a multi-stage prediction model is proposed to reduce the effects of overlap and misalignment caused by multiple class inequalities. The SMOTE uses the random sampling method and in the feature selection the wrapper and filter methods are used. The proposed model produces promising and comparable results, which are used to develop a performance model for predicting the unequal distribution