Explainable Temporal Machine Learning for Academic Risk Prediction and Student Intervention

Abstract

Predicting which students are likely to struggle is a well-studied problem in higher-education analytics, but what an institution should do with such predictions is less settled. This paper proposes a framework that treats academic risk as something that shifts over a semester rather than a label assigned once, and that closes the gap between predicting risk and acting on it. It combines longitudinal academic and behavioral indicators, temporal feature construction, machine-learning-based risk prediction, explainable AI, and a feedback loop that maps predictions to interventions and then checks whether those interventions changed anything. The framework builds on two existing studies covering complementary ground: one showing that dropout risk and its predictors shift month by month, and another showing that early grades and course history can forecast future performance and study duration. Rather than reporting new experimental results, the paper lays out a research methodology for temporal prediction, interpretability, intervention design, and post-intervention reassessment that a future empirical study could follow. Key Words:  academic risk prediction, explainable AI, higher education, machine learning, student intervention, temporal learning.