Abstract: OBJECTIVE To explore the application of the combined model of autoregressive integrated moving average (ARIMA) and Elman recurrent neural network (ERNN) (i.e., the ARIMA-ERNN model) in predicting the epidemic trend of hospital multi-drug resistant bacterial infections, and to provide evidence for hospitals to formulate prevention and control strategies for such infections. METHODS We collected the monthly infection cases of multidrug-resistant organisms (MDROs) at Guangzhou First People's Hospital from Jan. 2016 to Dec. 2025, with a total of 2 168 cases included in the analysis. R software version 4.4.2 was employed to develop the model, and autocorrelation function (ACF) and partial autocorrelation function (PACF) plots were utilized to determine the parameters of the ARIMA model and the Box-Ljung test to evaluate the model residuals. The ARIMA-ERNN model was screened based on mean absolute error (MAE), root mean squared error (RMSE) and mean absolute percentage error (MAPE). Data from Jan. 2025 to Dec. 2025 were selected as the test set to evaluate the model's predictive accuracy, and the dynamic trend of MDRO infections from Jan. 2026 to Jun. 2026 was predicted. RESULTS The incidence rate of MDRO infections from 2016 to 2025 exhibited dynamic changes (χ²=44.403, P<0.001), primarily occurring in the department of critical care medicine. The key bacteria involved were carbapenem-resistant Enterobacteriaceae, methicillin-resistant Staphylococcus aureus and carbapenem-resistant Acinetobacter baumannii. The optimal model, ARIMA(8,1,1), had an MAPE of 32.97%. The Box-Ljung test indicated no autocorrelation in the residuals. The optimal ARIMA-ERNN model had an iteration count of 6 900, a learning rate of 0.001, a momentum coefficient of 0.09 and a smoothing factor of 0.01. Both the actual and predicted values from Jan. 2025 to December 2025 showed a trend of initial increase followed by a decrease, with the model's predicted MAPE being 17.65%. The model predicted a low prevalence of MDRO infections from Jan. 2026 to June 2026, with the lowest incidence in February, followed by a rebound, and a slight increase from Apr. to Jun. CONCLUSION The ARIMA-ERNN model can be utilized for short-term prediction and dynamic analysis of the epidemic trend of hospital MDRO infections, providing technical support for early warning and prediction of such infections.