AI-BASED EARLY WARNING SYSTEM (EWS): PREDICTING DROPOUT AND LOW ACADEMIC PERFORMANCE RISK USING LEARNING ANALYTICS DATA

Abstract

This paper addresses the problem of early prediction of educational risk using learning analytics approaches. In distance education settings, a learner’s educational trajectory is reflected in digital traces recorded in an LMS; systematic analysis of these data enables early identification of the risk of dropout or low academic performance. The study aims to propose a data- and methodology-driven model of an Early Warning System (EWS) and to comparatively evaluate predictive models that classify at-risk learners based on engagement indicators from the first weeks of study. As research materials, an anonymized synthetic dataset (N=260) was used; it preserves the structure of LMS logs while preventing personal identification. Through feature engineering, metrics of participation, learning time, assessment performance, and social activity were constructed. Logistic regression, Random Forest, and Gradient Boosting were compared; evaluation metrics included ROC-AUC, accuracy, precision, recall, and F1. The results show that logistic regression achieved the highest discriminative performance (AUC=0.897; F1=0.774), demonstrating that learning risk can be assessed reliably at an early stage. The discussion substantiates the intervention design, ethical and pedagogical constraints, and approaches for integrating model explainability (XAI) into instructional decision-making.