A Hybrid Explainable Machine Learning Framework for Adaptive Recommendation in E-Learning Systems

Abstract

Personalized recommendation has become an essential component of modern e-learning systems, yet conventional collaborative- and content-based filtering approaches are often limited by data sparsity, insufficient contextual awareness, and limited interpretability. This paper proposes a hybrid explainable recommendation framework that integrates collaborative filtering, content-based filtering, bidirectional long short-term memory (BiLSTM) sequence modelling, reinforcement learning–based dynamic fusion, and SHAP-driven explainability to deliver adaptive and transparent learning recommendations. Unlike existing studies that primarily optimize either recommendation accuracy, adaptability, or interpretability, the proposed framework jointly addresses all three objectives within a unified architecture. The model was initially evaluated using a behaviourally motivated synthetic dataset comprising 250 learners and subsequently validated on the Open University Learning Analytics Dataset (OULAD). Comparative experiments included reproduced implementations of GRU4Rec, SASRec, and LightGCN under consistent evaluation settings. Experimental results demonstrate that the proposed approach achieves an accuracy of 92.8% while providing comprehensive feature-level explanations through SHAP analysis, thereby improving the transparency and trustworthiness of recommendations. Runtime evaluation and error analysis further confirm the practical feasibility of the framework. The findings indicate that combining sequential learning, adaptive fusion, and explainable artificial intelligence provides an effective balance between predictive performance, adaptability, and interpretability, making the proposed framework suitable for trustworthy intelligent learning environments.