Toward Personalized Adaptive Learning Using Artificial Intelligence and Physiological Signals

Personalized Adaptive Learning (PAL) systems aim to dynamically tailor instruction to learners’ cognitive and affective states, yet most existing approaches lack real-time, scalable indicators of learner status. Advances in wearable sensing and Internet of Things (IoT) technologies now enable unobtrusive monitoring of signals that reflect cognitive load, fatigue, and arousal, providing a foundation for physiology-informed PAL. This paper presents a PAL system that integrates real-time physiological deviations and accumulated fatigue into a performance update model, enabling session-wise modeling of learner performance. Based on this performance update model, we construct multiple difficulty adaptation policies. These policies range from rule-based heuristics to Artificial Intelligence (AI)-based strategies that optimize the PAL system. The AI-based methods leverage pseudo labels derived from physiological indicators and modeled performance dynamics, allowing supervised learning of adaptation policies without ground-truth performance data. Using the publicly available Psychological State Identification (PSI) dataset, we simulate multi-session learning scenarios and demonstrate that AI-based policies rapidly converge to near-perfect performance, consistently outperforming rule-based baselines.We also discuss key deployment challenges, including signal reliability, explainability, and the need for standardized learning analytics frameworks in real-world PAL systems