Forecasting influenza A and B positive case counts using a hybrid seasonal autoregressive integrated moving average and long short-term memory (SARIMA-LSTM) model: an epidemiological study from Jiande, Hangzhou, China

Influenza A and B impose substantial public-health burdens. Using hospital- based laboratory-confirmed surveillance data, this study aimed to characterise the epidemiological features of influenza A and B cases and forecast their temporal trends, so as to provide evidence for influenza interventions and epidemic preparedness among medical institutions in Jiande. Data for this descriptive study, consisting of laboratory-confirmed positive cases of influenza A and B spanning January 2019 to December 2024, were extracted from the medical-record registration system at the Jiande Branch of the Second Affiliated Hospital, Zhejiang University School of Medicine. Data were curated using Microsoft Excel 2016. A hybrid seasonal autoregressive integrated moving average-long short-term memory (SARIMA-LSTM) model was constructed in MATLAB R2024a to predict future trends in laboratory-confirmed positive case counts of influenza A and B. Laboratory-confirmed influenza A and B cases exhibited distinct seasonal patterns, with epidemic peaks predominantly occurring in winter and spring. The optimal SARIMA model was SARIMA (1, 0, 1)(2, 1, 1)₄ for influenza A and SARIMA (3, 0, 0)(2, 1, 1)₄ for influenza B. On the 2024 out-of-sample test set, compared with the standalone SARIMA model, the SARIMA-LSTM hybrid model reduced the MAE by 7.3% and RMSE by 7.1% for influenza A; for influenza B, MAE was reduced by 7.5% and RMSE by 7.5%. Compared with the standalone LSTM model, the hybrid model achieved a 9.4% reduction in MAE and a 4.6% reduction in RMSE for influenza A, alongside a 8.9% reduction in MAE and a 9.5% reduction in RMSE for influenza B. Notably, the SARIMA-LSTM hybrid model yielded better predictive performance for influenza B than for influenza A. Based on hospital-derived positive case counts, influenza A and B exhibited prominent winter-spring seasonal patterns. Compared with standalone LSTM and standalone SARIMA models, the SARIMA-LSTM hybrid model achieved better forecasting performance for both influenza A and B. This model can provide auxiliary references for the seasonal influenza epidemic trend analysis for local healthcare institutions and disease-control agencies. Not applicable.