Predicting Student Grade Point Average: Comparison of Machine Learning Regression Algorithms

Education data mining has emerged as a powerful technique for uncovering hidden patterns in educational data, forecasting academic achievement, and increasing retention rates. In this work, the performance of nine regression algorithms has been evaluated in predicting students’ academic success. Information from 650 students enrolled in three different computing majors has been assembled into a dataset. The following input attributes were chosen: attendance rate, course grade, gender, course category, delivery mode, school type, and high school score; the grade point average was the target variable. Findings indicate that Random Forest Regressor, Light Gradient Boosting Machine, Gradient Boosting Regressor, and Extra Tree are the four most effective regression algorithms in the order given. Except for the Light Gradient Boosting Machine approach, the other three algorithms showed that course grade is the most important predictor of a student's GPA, followed by high school score. All four algorithms showed that gender is the least reliable indicator of GPA. Future work will conduct sensitivity analysis to evaluate the impact of individual attributes on predictions to gain more insight into the factors affecting students’ performance. The effect of social, economic, and demographic features will also be investigated