Abstract
BACKGROUND: This study aimed to analyze the epidemiological characteristics of mumps in Ningbo, China, from 2005 to 2024, and to develop a predictive model using a Long Short-Term Memory (LSTM) neural network to forecast incidence trends. METHODS: Surveillance data of mumps cases were obtained from the China National Infectious Disease Network Direct Reporting System (CNIDNDRS). Descriptive statistics, Cochran-Armitage trend tests, and chi-square tests were employed to analyze epidemiological distributions. An LSTM model was constructed using monthly incidence data from January 2005 to December 2021 for training and validated with data from January 2022 to December 2024. Model generalizability was assessed using a synthetic dataset generated via time-series decomposition with stochastic noise. A SARIMA model was developed as a statistical baseline for comparison. RESULTS: A total of 40,444 mumps cases were reported in Ningbo during the study period, with an average annual incidence of 32.7 per 100,000 population. A significant decreasing trend was observed from 2005 to 2024 (Z = -98.6, P < 0.001). The disease exhibited a bimodal seasonal pattern (April–July and November–January) before 2020, which weakened thereafter. Incidence was significantly higher in males than females (1.67:1 ratio, P < 0.001), and the majority of cases occurred in children aged 3–14 years (83.0%) and students (49.8%). Spatial analysis revealed higher incidence in suburban counties compared to urban districts (P < 0.05). The LSTM model demonstrated strong predictive performance, with a mean absolute percentage error (MAPE) of 32.04% on test set, and maintained robustness on synthetic data. The LSTM outperformed the SARIMA model (test set MAPE = 40.22%). CONCLUSIONS: Mumps incidence in Ningbo has declined significantly, entering a sustained low-prevalence phase largely due to high vaccination coverage. The LSTM model effectively predicted incidence trends, particularly for epidemic peaks, and outperformed traditional time-series models.