Dynamic Learner Profiling Using Reinforcement Learning for Real-Time Adaptive Learning Environments

This study proposes a reinforcement learning–based dynamic learner profiling framework designed to enhance real-time adaptive learning in digital education environments. To address limitations of static and rule-based adaptive systems, the model integrates continuous event-log processing, feature engineering, and sequential policy optimization. The dataset consisted of 18,450 interaction logs collected from 300 simulated learning sessions, capturing behavioral, cognitive, and engagement indicators. Feature analysis revealed substantial variability in learner behavior, with response times ranging from 3.1 to 52.4 seconds (M = 16.8, SD = 8.2), hint usage frequencies from 0.00 to 0.88, and mastery scores spanning 0.12 to 0.95. The reinforcement learning agent was trained across 30,000 interaction steps, achieving stable convergence as indicated by a normalized reward increase of 0.73 over the first 300 episodes. Empirical results demonstrate that the adaptive RL policy substantially improved learner performance. Concept mastery increased from 0.54 to 0.72 (+33.3%), response time decreased from 16.8 to 12.4 seconds (–26.2%), and attempts per item were reduced by 22.5%. Engagement indicators improved markedly, with idle time dropping by 35.2% and hint frequency reduced by one-third. Analysis of action-selection behavior showed a balanced instructional strategy: increasing difficulty constituted 28% of actions, decreasing difficulty 22%, providing hints 19%, showing examples 17%, and delivering remedial material 14%. These distributions reveal a policy tuned to maintain optimal cognitive challenge while preventing overload. This work contributes a methodological advancement to the field of adaptive learning systems by integrating RL-driven decision-making with comprehensive state modeling, offering significant implications for intelligent tutoring systems, digital learning platforms, and data-driven instructional design.