Integrating Explainable Artificial Intelligence into Adaptive Learning Models for Transparent Student Feedback

This study proposes an explainability-centered adaptive learning framework that integrates Bayesian Knowledge Tracing (BKT), Gradient Boosting, and Explainable Artificial Intelligence (XAI) techniques to enhance transparency, trust, and performance in personalized learning systems. Using a dataset comprising 3,482 student interaction sequences, 22,910 log events, and five primary feature groups (Time_on_Task, Quiz_Score, Persistence_Index, Resource_Clicks, and Sequence_Position) the model was trained using a 70–15–15 stratified data split. The hybrid BKT–Boosting model achieved a mastery-prediction accuracy of 0.87, an AUC of 0.91, and a recommendation precision of 0.86 for medium-difficulty tasks. Error distribution analysis revealed a tight range (SD = 0.12) with minimal outliers, demonstrating consistent model reliability. Overall, findings demonstrate that embedding XAI directly into adaptive learning pipelines improves not only model interpretability but also learner engagement, comprehension, and mastery progression. The results strongly support explainability as a foundational component of modern AI-driven educational systems rather than an optional add-on. This research provides an integrated methodological and empirical basis for developing transparent, trustworthy, and pedagogically coherent adaptive learning systems at scale.