Comparative study of Bayesian and Frequentist methods for epidemic forecasting: Insights from simulated and historical data

Abstract

Accurate epidemic forecasting is critical for effective public health interventions. This study compares Bayesian and Frequentist estimation frameworks within deterministic compartmental epidemic models, focusing on nonlinear least squares optimization versus Bayesian inference using MCMC sampling via Stan. We compare forecasting performance under shared modeling structure and error assumptions for specific implementations of both approaches. We assess performance on simulated datasets (with R0 values of 2 and 1.5) and historical datasets including the 1918 influenza pandemic, 1896-97 Bombay plague, and COVID-19 pandemic. Evaluation metrics include Mean Absolute Error, Root Mean Squared Error, Weighted Interval Score, and 95% prediction interval coverage. Forecasting performance depends on epidemic phase and dataset characteristics, with no method consistently outperforming across all contexts. Frequentist methods perform well at peak and post-peak phases but are less accurate pre-peak. Bayesian methods, particularly with uniform priors, offer better early-epidemic accuracy and stronger uncertainty quantification, especially valuable when data are sparse or noisy. Frequentist methods often yield more accurate point forecasts with lower error metrics, though their interval estimates may be less robust. We examine how prior choice influences Bayesian forecasts and how extending forecasting horizons affects convergence. These findings offer practical guidance for choosing estimation strategies tailored to epidemic phase and data quality, supporting more effective public health interventions.