Answers are grounded in this paper and cite their sources.
Comparative study evaluates forecasting models for hepatitis B cases in China, indicating model selection implications.
BACKGROUND: Robust forecasting of hepatitis B trends is important for long-term surveillance and public health planning in China. However, evidence remains limited regarding how different forecasting frameworks perform when applied to a long-term national hepatitis B surveillance series under a unified out-of-sample evaluation design. METHODS: Monthly reported hepatitis B cases in China from January 2004 to December 2024 were obtained from national public surveillance databases. Data from January 2004 to December 2023 were used as the training set, whereas data from January to December 2024 were reserved as an independent validation set. Ten forecasting models were evaluated, including Seasonal autoregressive integrated moving average (SARIMA), Bayesian structural time series (BSTS), four standalone machine-learning models (random forest, support vector machine, XGBoost, and LightGBM), and four residual-based SARIMA-machine-learning hybrid models. Model performance was assessed using mean absolute error (MAE), range-normalized root mean square error (NRMSE), mean absolute percentage error (MAPE) and Nash-Sutcliffe Efficiency (NSE). In addition, the average annual percentage change (AAPC) was estimated to summarize the long-term annual trend in reported case counts.The best-performing models were further used to forecast monthly hepatitis B case counts in 2025. RESULTS: was selected as the optimal model. In the overall model comparison, BSTS achieved the best validation performance, with the lowest MAE (10014.27), NRMSE (38.49%), MAPE (7.18%) and highest NSE(-0.92) among all ten models. Among the hybrid models, SARIMA-LightGBM performed best, with corresponding validation values of 10903.35, 40.14%, 8.07% and - 1.09, respectively. In 2025, BSTS achieved an MAE of 13,437.42, NRMSE of 9.53%, MAPE of 12.04%, and NSE of - 0.41, with 95% posterior predictive intervals used to quantify forecast uncertainty. CONCLUSIONS: Forecasting performance differed across model classes in long-term national hepatitis B surveillance. BSTS showed the best overall predictive performance. These findings support empirical model selection and suggest that BSTS-based forecasting may serve as a supportive tool for early warning, resource allocation, and long-term hepatitis B surveillance planning when interpreted together with reporting-system and policy context.
No takes yet. Share an insight, caveat, or question.
Lu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: