Explainable AI for Personalized Learning: Enhancing Transparency in Adaptive Education Systems

This research focuses on the integration of Explainable Artificial Intelligence (XAI) into adaptive learning systems to enhance transparency, trust, and decision-making for educators and learners. To address the shortcomings of black-box personalized learning models, this research used an XAIintegrated prototype in three higher education institutions (HEIs) in Manila for a period of four weeks, engaging 587 participants (43 professors and 544 students). The adaptive learning system used a gradient boosted recommender model and an early warning classifier, integrated with a multi-method XAI explanation pipeline using TreeSHAP, LIME, and Anchors. Findings showed strong and significant gains in a variety of areas. Teacher trust levels were boosted by +0.72 points (3.28 to 4.00), and student trust levels by +0.62 points (3.08 to 3.70), both $p\lt.001$. Engagement with learning increased, with active engagement each week improving by 5.3 percentage points and inactive weeks falling by 4.5 points. Academic achievement also improved, with module test results rising by 4.2 points on average. Views of data protection increased by +0.78, indicating increased confidence in privacyprotecting systems. The early warning system model obtained an AUC of 0.812, and federated learning maintained a nearequivalent level of accuracy compared to the centralized approach (0.874 vs. 0.878). The results obtained from this study suggest that the application of XAI-based personalization can significantly improve the interpretability, trustworthiness, and learning outcomes while promoting ethical and privacy-conscious AI adoption in the education sector.