STUDENT MENTAL FITNESS AND ACADEMIC PERFORMANCE ANALYSIS USING MLR AND RANDOM FOREST ALGORITHM

Abstract

Predicting student academic performance is a critical task for educational institutions seeking to provide timely interventions. This paper presents a machine learning framework to predict student letter grades (A–F) based on academic and mental-fitness features derived from 5,000 student records. Two classifiers—Multinomial Logistic Regression (MLR) and Random Forest (RF)—are developed, evaluated, and compared. Feature engineering introduces two composite proxies, Motivation_Score and Focus_Score, to capture psychological influences on learning. SHAP (SHapley Additive exPlanations) provides post-hoc interpretability for both models. MLR achieves 92.2% accuracy and a weighted F1-score of 0.927, outperforming RF (79.1% accuracy, weighted F1 of 0.759) under balanced class weighting. An interactive Gradio application delivers real-time grade predictions with feature-level explanations. Results show that Final Score, Assignments Average, and Quiz Average are dominant predictors, while mental-fitness proxies contribute meaningfully for borderline students.