Deep reinforcement learning based personalized mathematics learning path planning algorithm for children

Early childhood mathematics learning is increasingly supported by adaptive technologies; however, most current systems depend on static or rule-based personalization, which restricts their ability to effectively adapt to changing learner behavior. Conventional adaptive learning approaches do not effectively integrate multimodal behavioral and emotional signals, leading to reduced engagement, slower knowledge acquisition, and limited personalization for preschool learners. This research proposes a Learning-Based Q Network (LBQN) to design a personalized mathematics learning path that adapts dynamically to children’s cognitive and behavioral states. Personalized Math Learning Behavior Dataset was utilized (4000 rows and 14 columns). Preprocessing involves median filtering for noise reduction, min–max normalization for feature scaling, and temporal alignment to synchronize multimodal behavioral using sliding window. Feature extraction was conducted using statistical features (mean, variance) and behavioral indicators such as response time and error rate. The proposed LBQN method integrates Double Deep Q-Network (DDQN) with Teaching-Learning-Based Optimization (TLBO) to optimize action selection and learning path planning through reward-driven policy updates implemented in Python. Experimental results demonstrate that the proposed LBQN model outperformed existing A-DRL, MLP-LSTM, and RL-DKT by achieving an accuracy of 93.8%, with recall (90.6%), F1-score (91.7%), and user satisfaction rating (4.9/5). The proposed LBQN method effectively enhances personalized mathematics learning by combining DDQN-based reinforcement learning with offline Teaching–Learning-Based Optimization (TLBO) for parameter refinement. This sequential hybrid framework improves convergence stability, learning path quality, and adaptive data-driven decision making for young children. This provides an effective foundation for intelligent tutoring systems by enabling data-driven personalized mathematics instruction, improving learning outcomes, and supporting adaptive decision-making in early childhood education.