Reinforcement Learning-Based Dynamic Fuzzy Weight Adjustment for Adaptive User Interfaces in Educational Software

Adaptive educational systems are essential for addressing the diverse learning needs of students by dynamically adjusting instructional content and user interfaces (UI) based on real-time performance. Traditional adaptive learning environments often rely on static fuzzy logic rules, which lack the flexibility to evolve with learners’ changing behaviors. To address this limitation, this paper presents an adaptive UI system for educational software in Java programming, integrating fuzzy logic and reinforcement learning (RL) to personalize learning experiences. The system consists of two main modules: (a) the Fuzzy Inference Module, which classifies learners into Fast, Moderate, or Slow categories based on triangular membership functions, and (b) the Reinforcement Learning Optimization Module, which dynamically adjusts the fuzzy membership function thresholds to enhance personalization over time. By refining the timing and necessity of UI modifications, the system optimizes hints, difficulty levels, and structured guidance, ensuring interventions are neither premature nor delayed. The system was evaluated in educational software for Java programming, with 100 postgraduate students. The evaluation, based on learning efficiency, engagement, and usability metrics, demonstrated promising results, particularly for slow and moderate learners, confirming that reinforcement learning-driven fuzzy weight adjustments significantly improve adaptive UI effectiveness.