A Reinforcement Learning Approach to Personalized Asthma Exacerbation Prediction Using Proximal Policy Optimization

Asthma exacerbation prediction is critical for preventing severe respiratory complications and improving patient outcomes. Traditional predictive models rely on static machine learning approaches, which lack adaptability to evolving patient conditions and environmental changes. This study proposes a reinforcement learning (RL)-based model using Proximal Policy Optimization (PPO) to enhance personalized early prediction of asthma exacerbation attacks. Unlike conventional models, PPO dynamically refines its policy through iterative learning, allowing for real-time, adaptive risk assessment based on personalized patient data and environmental factors. The PPO model was trained on a comprehensive asthma dataset, integrating demographic, environmental, lifestyle, and Asthma Control Test (ACT) scores. The model achieved 96.60% accuracy, 95.79% precision, 96.65% recall, and 95.92% F1-score, outperforming baseline RL algorithms such as Deep Q-Learning (92.21% accuracy), Advantage Actor-Critic (94.34%), and Trust Region Policy Optimization (95.12%). Comparative evaluation with existing studies further demonstrated PPO’s superior recall and sensitivity to exacerbation cases, making it highly suitable for personalized asthma management. RL performance metrics confirmed the model’s stability and convergence, including cumulative rewards, policy, and value loss. These findings highlight PPO’s potential for real-time clinical integration, enabling continuous patient monitoring and proactive intervention strategies. Future work will focus on training the model on larger, multi-source datasets to improve generalization across diverse populations. Additionally, further research will explore deployment in telemedicine, integration with wearable health technologies, and adaptability to other chronic diseases, ensuring scalability and real-world applicability in personalized healthcare systems.