Student Performance Prediction Using Machine Learning: A Comparative Analysis of Decision Tree, Random Forest, and Optimized XGBoost

Abstract

Student performance prediction has become an important application of Machine Learning in educational data mining, enabling institutions to identify academically weak students at an early stage and provide timely academic support. Accurate prediction of student performance helps educators implement personalized learning strategies, reduce failure rates, and improve overall educational outcomes. This paper presents a Machine Learning-based Student Performance Prediction System that compares the performance of two existing classification algorithms, Decision Tree and Random Forest, with a proposed Optimized XGBoost algorithm. The proposed system utilizes the UCI Student Performance dataset, which combines Mathematics and Portuguese student records into a single dataset containing 1,044 student instances. The dataset undergoes comprehensive preprocessing, including duplicate removal, categorical encoding, feature scaling, feature selection, and performance category generation. Three supervised Machine Learning models are trained and evaluated using Accuracy, Precision, Recall, F1-Score, Log Loss, Cohen's Kappa, and Matthews Correlation Coefficient (MCC). Experimental results demonstrate that the proposed Optimized XGBoost model achieves the highest prediction accuracy of 67.94%, outperforming Random Forest (65.55%) and Decision Tree (61.72%). The trained model is deployed using the Streamlit framework to provide a user-friendly web application capable of predicting student performance and generating academic recommendations in real time. The proposed system offers an efficient decision-support tool for educational institutions, enabling proactive academic intervention and data-driven educational management.