Causality-Aware Deep Learning for Climate-Sensitive Infectious Disease Forecasting via Causal Graphs and Counterfactual Simulation

Accurately forecasting infectious disease outbreaks under accelerating climate change presents a profound challenge that demands both predictive accuracy and mechanistic interpretability. This study introduces a novel Causality-Aware Deep Learning (CADL) framework that systematically integrates causal structure learning, deep temporal forecasting, counterfactual simulation, and explainability into a unified architecture for forecasting climate-sensitive infectious diseases such as dengue fever. By combining directed acyclic graph discovery, hybrid graph neural networks and transformers, and structural causal models, the proposed framework simultaneously delivers high forecasting accuracy and actionable causal insights. Theoretical analysis establishes stability and convergence guarantees, while synthetic experiments demonstrate superior forecasting performance — achieving a 26.8% reduction in mean absolute error compared to standard deep learning baselines including LSTM, Transformer, and GNN models. The framework further enables counterfactual forecasting under alternative climate change scenarios (e.g., RCP 4.5 and RCP 8.5), providing policymakers with scenario-based risk quantification to support adaptive public health strategies. CADL represents a significant advance in explainable AI for epidemic forecasting, with broad implications for climate-resilient global health preparedness.