Early GPA Prediction: A Comparison of Fuzzy Inference, Regression, and Decision Trees

Abstract

Student dropout remains a persistent challenge in higher education, making the early identification of at-risk students essential for timely intervention. In the Informatics study program examined in this research, the first-stage academic evaluation is conducted at the end of the fourth semester, creating a need for predictions that can be made well before this milestone. This study compares a Mamdani Fuzzy Inference System (FIS), linear regression, and decision tree in predicting fourth-semester GPA based on students' grades in three foundational first-semester courses: Basic Mathematics, Discrete Mathematics, and Introduction to Programming. The dataset consists of the academic records of 254 students across four cohorts. The Mamdani FIS rule base was constructed in a data-driven manner, while linear regression and decision tree served as comparative baselines. All three models were evaluated using 5-fold cross-validation, with Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) as evaluation metrics. Linear regression achieved higher accuracy and greater stability, with an average RMSE of 0.3766 and MAE of 0.2874, compared to the Mamdani FIS which obtained an RMSE of 0.5052 and MAE of 0.3660. The decision tree (RMSE = 0.4776, MAE = 0.3558) ranked between them and was the most stable. However, the Mamdani FIS produced an interpretable linguistic rule base, of which the majority remained consistent across folds, offering transparency that supports decision-making in identifying at-risk students. These findings highlight a trade-off between predictive accuracy and interpretability, indicating that the choice of method depends on whether the priority is prediction accuracy or model transparency.