Prediction of Daily Outpatient Volume Using Ensemble Empirical Mode Decomposition and Long-Short Term Memory During the Incident

Effective hospital resource management is crucial for service cost and workforce efficiency, especially during abnormal situations with sudden changes in patient volumes. Hospitals often face challenges in resource management and find it difficult to predict patient numbers due to uncertain and volatile data. If a hospital can accurately predict the number of patients during abnormal situations, it can allocate resources to align with the expected patient influx. This helps the hospital maintain appropriate costs and optimize workforce efficiency. In this research, we propose a solution to predict the number of daily outpatient visits from a private hospital in Thailand during the COVID-19 outbreak period as a case study. We use the Ensemble Empirical Mode Decomposition (EEMD) to preprocess the data by decomposing it into intrinsic mode functions, reducing the impact of data volatility and predict these intrinsic mode functions using a Long Short-Term Memory (LSTM) model with the service utilization pattern (seasonal value) as a lookback period to enhance prediction efficiency. This approach is referred to as the EEMD-LSTM model. Experimental results indicate that, compared to LSTM, ARIMA, and AutoML models evaluated for prediction periods of 7, 14, 21, and 28 days using metrics including MAE, RMSE, MAPE, and MAE Ratio, the EEMD-LSTM model demonstrates superior performance in predicting patient numbers during abnormal events. Unlike other models, the EEMD-LSTM model exhibits minimal changes in error when the prediction period is adjusted, highlighting its robustness. Overall, the EEMD-LSTM model consistently outperforms all other models assessed. When applied to data from the orthopedic department, although the MAE values are similar, noticeable differences in trends are observed. Moreover, when predicting data from the pediatric department, the EEMD-LSTM model proves to be more effective than all other models. This model can be effectively utilized by hospitals for resource management and planning, including medication and staffing, ensuring optimal service delivery to patients during abnormal situations.