Student Academic Performance Prediction using Machine Learning with Various Features and Scenarios

This paper examines the use of machine learning algorithms to predict student performance, focusing on past academic achievements and scores from core subjects in sciencemath program of high school. The study addresses two prediction types: regression for predicting GPA and classification for predicting grades. Experiments were conducted on 853 students, implementing predictive models across five scenarios corresponding to five semesters. Linear Regression achieved the highest R-squared value of 0.8911. For grade classification, Support Vector Machine, K-Nearest Neighbors, Logistic Regression, and Artificial Neural Networks with Grid-search hyper parameters performed better than default hyper-parameters. Random Forest and K-Nearest Neighbor achieved 100% accuracy in grade predictions across all scenarios. Achieving perfect accuracy indicates that these models could play a transformative role in educational settings by providing precise predictions that could enhance personalized learning strategies and academic support systems.