Regression performance metrics for three Bayesian NN models for four forecast horizons (γ = 7, 14, 21, and 28 days ahead).

Negative log likelihood (NLL), continuous ranked probability score (CRPS) and Skill compare the accuracy weighted by the uncertainty of forecasts. MAE is the mean absolute error, and r is the bivariate correlation between forecasts and reported ILI rates. Best results for each metric and forecast horizon are shown in bold. The last three columns are performances averaged over the four test flu seasons (from 2015/16 to 2018/19).