ADAPTIVE DEEP REINFORCEMENT LEARNING FOR PERSONALIZED COGNITIVE BEHAVIORAL THERAPY VIA MOBILE HEALTH PLATFORMS

Abstract

Cognitive Behavioral Therapy (CBT) is a widely recognized psychological intervention for managing various mental health disorders. However, the effectiveness of CBT often depends on personalized delivery tailored to an individual's unique needs, which is challenging in traditional settings. Conventional mobile health (mHealth) platforms for CBT lack adaptive personalization capabilities, leading to suboptimal engagement and treatment outcomes. There is a need for intelligent systems that can dynamically adapt CBT content and interventions based on user feedback and behavioral data. This study proposes an adaptive deep reinforcement learning (DRL) framework that personalizes CBT interventions delivered through mHealth platforms. The DRL agent models user states using multisource data (behavioral, psychological assessments, and interaction logs) and learns an optimal policy to recommend tailored CBT activities. The framework employs a deep Q-network (DQN) with experience replay and target networks to stabilize training, incorporating user feedback as rewards to improve personalization over time. Experiments on simulated and real-world user data show that the proposed DRL-based system significantly improves user engagement, adherence to CBT protocols, and symptom reduction compared to static recommendation baselines. The system effectively adapts to changing user states and optimizes treatment strategies, validating its potential for scalable personalized mental health interventions.