Authors: Dr Ansari Pulickal Abdul Azeez, Farooq Sajjad
Abstract: Digitization of business education at an unprecedented rate has made available to educators large amounts of student interaction data that can inform data-driven learning interventions. In this paper, we propose an Artificial Intelligence-driven Learning Analytics (AI-LA) system architecture, which incorporates multi-stream data sources (Learning Management System (LMS) logs, clickstream analysis, test/assignment submissions, and engagement data) to model, explain, and improve student engagement and performance. Our approach leverages a novel combination of techniques that include a Temporal Fusion Transformer (TFT) model for sequential behavior prediction, SHapley Additive exPlanations (SHAP) for interpretable feature importance, and reinforcement learning (RL) engine for personalized intervention recommendations. Our model was tested using longitudinal data from 3,400+ business management students in 24 courses over three academic years (2022-2025). It predicted at-risk students with up to 89.5% accuracy si
DOI: https://doi.org/10.5281/zenodo.20700051
