Uncovering Key Influences on Student Performance Through Educational Data Mining: An XGBoost Approach with Cluster Analysis

This paper presents a machine learning framework achieving 97.7% accuracy (R 2 = 0.977) in predicting student performance by integrating academic metrics (e.g., exam scores) with behavioral indicators (question-asking frequency, ChatGPT usage). K-means clustering reveals three distinct student groups with significant performance gaps (49.33 vs. 40.05 average marks). Deployed via a Streamlit interface, the system demon- strates that behavioral features contribute 19.7% additional explanatory power beyond traditional academic data.