Implementation of Machine Learning Algorithms for Predicting Student Final GPA Using Multiclass Classification Models

Sari

This study aims to predict students’ final Grade Point Average (GPA) and study duration by applying the Random Forest Regressor algorithm based on historical academic data. The dataset includes key variables such as semester GPA, socio-economic background, demographic information, study habits, and the complexity level of the enrolled courses. The regression analysis results indicate that the model's performance was suboptimal, with a Mean Squared Error (MSE) of 0.341 for GPA prediction and 3.831 for study duration estimation. Additionally, the negative R-squared (R²) values reflect the model's limited ability to explain data variability. As an alternative, a multi-class classification approach was implemented to categorize students into final GPA groups, including Cum Laude, Very Satisfactory, Satisfactory, and Adequate. At this stage, the model achieved a remarkably high accuracy of 99.8% with an error rate of only 0.03. These findings demonstrate that the classification approach is more effective than regression in predicting academic performance. This research contributes to the development of data-driven academic decision support systems. Future studies are recommended to explore feature optimization techniques and alternative algorithms to enhance overall prediction performance.