Q-Learning for Personalized Adaptive Hint Timing in Game Interventions for Children with Autism Spectrum Disorder: A Feasibility Study

This study demonstrates the feasibility of employing a Q-learning model to personalize hint delay times in digital learning tasks for children with autism spectrum disorder (ASD). Leveraging data from a previous experiment, the model was developed to address the wide variability in ASD learning profiles. K-means clustering classified performance metrics (e.g., accuracy, completion time) into distinct learning profiles, which defined the Q-learning states. The model dynamically adjusted hint delays based on each child's real-time performance, ensuring individualized support aligned with their cognitive and motor abilities. Simulation results confirmed the model's effective adaptation across learning groups: fast learners (average accuracy 85.98%) received shorter delays to maintain engagement, while moderate learners (average accuracy 74.70%) received optimal consistent delays, and slow learners (average accuracy 59.36%) were given longer delays to support task completion. The converged Q-table values consistently reflected these tailored adjustments, highlighting the model's potential for real-time personalization. This adaptive approach shows promise for enhancing sustained attention, working memory, processing speed, and motor control in children with ASD.