Human–Computer Interaction Patterns in E-Learning: Insights from Learning Behavior Mining for Performance Prediction

Abstract

Human--computer interaction (HCI) plays a significant role in shaping learner engagement and outcomes in e-learning environments. Existing learning analytics approaches often rely on aggregated behavioral features, thereby limiting interpretability and weakening the connection between prediction and learner interaction processes. This study proposes an HCI-informed learning behavior representation that captures navigation dynamics, temporal interaction characteristics, engagement signals, and self-regulation behaviors from e-learning interaction logs. Using the Open University Learning Analytics Dataset (OULAD), logistic regression, random forest, and gradient boosting models were evaluated under static and HCI-informed feature settings. The proposed HCI-informed representation consistently improved predictive performance across all evaluated models. The best-performing configuration, Random Forest with HCI-informed features, achieved an accuracy of 0.858, an F1-score of 0.872, and an AUC of 0.925, and AUC of 0.925, compared with 0.796, 0.810, and 0.878 respectively for the static baseline. Statistical validation further demonstrated significant AUC improvements across all models (p < 0.001). Beyond prediction, the proposed representation provides interpretable evidence linking navigation regularity, temporal engagement, and self-regulation patterns with learner outcomes. These findings demonstrate that learning analytics can become more explanatory and actionable when interaction behavior is modeled from an HCI-informed perspective.