EMOTION-AWARE REINFORCEMENT LEARNING FOR ADAPTIVE VIDEO GAMING THERAPY

Abstract

Reinforcement Learning (RL) in Video Gaming Therapy, combined with emotion detection, dynamically adapts game environments to optimize therapeutic outcomes by personalizing experiences based on real-time emotional responses.A key challenge is integrating accurate emotion detection with reinforcement learning to personalize video game therapy effectively while ensuring real-time responsiveness and user engagement.This model presents a novel framework for an Inverted Residual Adaptive Convo-Depth NeuroNet that integrates Emotion Detection and Reinforcement Learning (RL) to create a therapy system for an adaptive video gaming therapy system. The Emotion Detection Module uses a pre-trained MobileNetV2 to analyzethe data and classify the emotional state, such as Happy, Fear, Angry, Disgust, Neutral, Sad, and Surprise. The Reinforcement Learning Module leverages Deep Q-Network (DQN)algorithms, where the state combines the player's emotional state and game performance. The RL model adjusts the game difficulty and provides personalized rewards. This methodpurposes to create a more immersive and effective video game therapy environment, where the game evolves in response to both emotional and performance rewards, ensuring a personalized and engaging experience.The Python-based simulation achieved performance metrics ranging from 83% to 85%, including F1-score, precision, recall, and accuracy, validating the effectiveness of the emotion-aware adaptive therapy system.