Q-learning Adaptive Mathematics Game for Early Numeracy Learning Evaluation

Early numeracy learning in elementary school requires learning media that can respond to differences in students’ abilities, especially when students show different levels of accuracy, response time, help use, and task completion. This study evaluated a Q-learning-based adaptive mathematics game designed to support early numeracy learning among first-grade elementary students. The evaluation focused on pretest–posttest learning outcome changes and adaptive model performance. The study involved 38 first-grade students from SDN 1 Langseb and used a one-group pretest-posttest design. Students completed a pretest, played a six-level adaptive mathematics game, completed a posttest, and were observed by the teacher and researcher during implementation. Learning outcome changes were analyzed using gain, N-Gain, and the Wilcoxon Signed-Rank Test. Adaptive model performance was evaluated through student states, adaptive actions, reward behavior, Q-value update, difficulty transition, decision consistency, and classroom feasibility. The results showed that the mean score increased from 43.95 in the pretest to 80.79 in the posttest, with a mean N-Gain of 0.666. The Wilcoxon test indicated a significant difference between pretest and posttest scores, with p < 0.001. The model produced 228 adaptive decisions across six levels, with Increase, Maintain, and Decrease/Reinforcement actions. The reward pattern was consistent with the adaptive logic, and the mean Q-value increased from 1.115 to 1.199. The findings show improved early-numeracy scores after the intervention and interpretable adaptive difficulty behavior in the observed classroom context. Because the study used a one-group pretest–posttest design without a comparison group, the observed improvement cannot be attributed solely to the adaptive game.