Sequential Pattern Mining for Predicting Student Dropout Risk From Learning Management System Interaction Logs

The increasing availability of Learning Management System (LMS) interaction logs enables early identification of students at risk of dropout. However, conventional prediction approaches often do not consider the temporal relationships among sequential learning activities. In this study, we propose a sequential pattern mining approach for dropout risk prediction (SPM-DRP) to identify behavioral sequences and predict dropout risk from LMS interaction logs. The proposed method organizes student activities sequentially, uses the PrefixSpan algorithm to identify frequent and dropout-associated patterns, and finally transforms discriminative patterns into predictive features for dropout classification. The dataset includes 120,542 enrollment records and 8,157,277 learning behavior records, with labels indicating dropout events and corresponding event data from seven LMS types. We measure performance using Lift, Pattern Growth Rate (PGR), PR-AUC, Brier Score (BS), and Matthews Correlation Coefficient (MCC), with the proposed SPM-DRP achieving values of 2.68, 33.2%, 0.914, 0.087, and 0.846, respectively. The findings illustrate the potential of the sequential behavioral patterns to capture dropout-related learning patterns and to assist in predicting dropout. The proposed SPM-DRP framework offers an actionable, interpretable method to identify deviations in student engagement and support timely academic intervention and retention strategies.