An Explainable Deep Reinforcement Learning Approach for Intelligent Resource Allocation and Epidemic Preparedness in Public Healthcare Systems

Abstract

Epidemics often lead to poor, delayed, and uneven distribution of public health resources, thereby pushing public healthcare systems to the limits of their capacities. This research introduces a "beds, healthcare personnel, testing kits, vaccines, medicines, and emergency funds" distribution model using explainable deep reinforcement learning for 20 public healthcare facilities that serves 50,000 adolescents and includes realistic 14-day demand forecasting using an LSTM network. Epidemiological, operational and demographic drivers of allocation decisions are captured through SHAP explanations, and constraints on capacity, budget, emergency-reserve and geographic-equity ensure transparent planning. The analyses of performance are done based on demand satisfaction, resource utilization, preparedness, unmet critical demand, vaccine wastage, epidemic peak demand, waiting time, rural–urban service disparity, explanation fidelity. The following reference values are considered: 92% satisfied with the demand; 90% efficiency of the resources; 85/100 preparedness; 5% of unmet critical demand; 3% of vaccine wastage; and 90% explanation fidelity. The method is conducive to efficient, equitable and interpretable epidemic-preparedness decision making.