2026 International Conference on Networks Computers and Communications (ICNCCom 2026)

The importance of quickly identifying at-risk students for online learning is paramount when providing support through timely interventions. This paper outlines an explainable predictive analytics system for predicting a three-class outcome, Pass, Distinction, and At Risk, for students enrolled in an organisation via the Open University Learning Analytics Dataset based solely on course activity during the first 6 weeks of the course. An educator-actionable definition of At Risk merges Fail and Withdrawn students because they will require similar intervention at approximately the same time. A classifier built using XGBoost with 16 behavioral, assessment and demographic features achieved 67.90% accuracy on the stratified holdout test dataset. The full model has a 67.94% mean accuracy when evaluated through 5-fold cross-validation and has improved accuracy from the 9 feature baseline of 66.13% to the full 16 feature model. SHAP explanations enable per-student interpretability and support the automatic generation of personalized learning recommendations based on feature attributions. The final outcome of this work indicates that timely and targeted interventions by educators can occur through competitive and transparent early predictions being made before the midpoint of the course.

Learning analytics,SHAP,XGBoost