A Longitudinal Study of Student Performance Prediction Models Using Educational Data Mining and Behavioral Analytics

Abstract

This study explores the development and performance of student performance prediction models through a longitudinal lens, utilizing educational data mining (EDM) and behavioral analytics. As institutions transition to hybrid and digital learning models, predicting academic outcomes becomes crucial for early intervention and personalized support. We evaluate a range of machine learning models trained on multi-year student data, including log-ins, submission timestamps, and interaction frequencies with learning management systems (LMSs). The findings suggest that behavioral features—particularly time-on-task and consistency of engagement are strong predictors of academic success. Our study also highlights how predictive accuracy evolves over time, emphasizing the importance of temporal dynamics in educational analytics.