Benchmarking Counterfactual Prediction in Epidemic Time Series with Time-Varying Interventions
Researchers have developed a new benchmark for evaluating deep learning models in predicting epidemic trajectories under dynamic interventions. This benchmark addresses the limitations of existing datasets by providing realistic counterfactual outcomes, supporting both static and time-varying treatments. It utilizes an agent-based model calibrated with real-world data to generate trajectories for over 150 U.S. counties, enabling a comprehensive assessment of causal inference methods. AI
IMPACT Provides a robust evaluation framework for AI models in public health, potentially improving epidemic forecasting and intervention strategies.