Using Explainable Artificial Intelligence to Predict Student Dropout Risk from Learning Management System Behavioural Data Early

Abstract

Student attrition remains a persistent challenge in higher education, affecting institutional performance, educational equity, resource allocation, and learners’ long-term academic outcomes. The expansion of digital learning environments has created extensive behavioural data that can reveal changes in student engagement before conventional academic indicators identify withdrawal risk. This study proposes an explainable artificial intelligence framework for early prediction of student dropout using longitudinal behavioural data captured through Learning Management Systems. The framework integrates login frequency, session regularity, content access, assessment submissions, discussion participation, inactivity duration, learning-resource consumption, and temporal engagement changes to construct dynamic student-risk profiles. Machine-learning models are developed to estimate dropout probability at successive stages of the academic period, while explainability techniques identify the behavioural factors contributing to individual predictions. Particular emphasis is placed on distinguishing temporary disengagement from persistent behavioural deterioration and determining how early reliable warnings can be generated. Model performance is evaluated using discrimination, calibration, class-imbalance-sensitive metrics, prediction lead time, and explanation consistency. The proposed approach supports transparent early-warning systems that enable institutions to identify emerging disengagement patterns and implement timely, evidence-informed student support interventions.