DEVELOPMENT OF STUDENT ACADEMIC PERFORMANCE PREDICTION SYSTEM

ABSTRACT This study was centred around developing a straightforward and smart system to assess a student’s potential academic performance using actual historical academic data. The dataset came from Kaggle and had the records of students from two Portuguese secondary schools. Unlike many other datasets which only indicate whether a student passed or failed, this dataset also included actual final grades which enabled us to use regression as opposed to mere classification. To enhance the dataset for predictions aligned with university standards, we selected the most relevant characteristics including term grades, absences, study time, health, and converted the grades to a 5.0 GPA scale because that’s what most universities use. We then trained a machine learning model we selected called Random Forest Regressor. The model performed remarkably well in predicting student grades with an accuracy score (R²) of 0.819, which improved to 0.8377 after model tuning. Additionally, through analysis, it emerged that the best predictor of student performance for the next term was their previous term grade (G2). While the dataset was valuable in enabling us to build a working system, we believe collecting data directly from universities as well as incorporating other school-related factors would significantly improve the system in future versions. This would improve the system's performance by making it more accurate and useful in real school environments.