Academic Temporal Pattern Mining for Enhanced Student Performance Prediction: A Hybrid Machine Learning Approach

Predicting student academic performance effectively remains challenging due to the dynamic nature of learning behavior and the complex interplay of factors affecting student success. This paper introduces Academic Temporal Pattern Mining (ATPM), a novel approach that identifies critical time-based patterns in student academic behavior to enhance performance prediction accuracy. Unlike traditional methods that rely on static snapshots of student data, our approach analyzes sequential patterns in study habits, submission timings, and performance trends. Using a UCI Student Performance Dataset of 1,000 undergraduates, we demonstrate that temporal patterns such as assignment submission sequences and grade progression provide crucial predictive insights. Our hybrid model, combining Random Forest for feature stability and XGBoost for capturing non-linear relationships, achieves an accuracy of 87%, representing a 15% improvement over conventional regression-based methods. Moreover, we rigorously evaluate the statistical significance of these gains (p < 0.05) and show an 80% precision threshold for early warning up to three assessment cycles before final grades. This research contributes to educational data mining by offering a practical and scalable framework for temporal pattern analysis that leverages readily available academic data for timely, actionable interventions.