Predicting daily patient arrivals in the emergency department based on a deep neural network approach: A Case Study

The goal of this study is to analyze the performance of four forecasting models in predicting daily patient arrivals in the emergency department (ED). Due to the fact that emergency patient flow is highly uncertain and dynamic, this forecasting problem is a challenging task. We evaluated different time series models to forecast ED daily patient arrivals at General Hospital of Cu Chi Area in Ho Chi Minh city, Vietnam. The forecasting models tested in this work are seasonal multiplicative Holt-Winters (HW), seasonal artificial neural network (SANN), the hybrid method which combines Holt-Winters with SANN and the deep neural network model: Long Short Term Memory (LSTM) network. The experimental results show that all the four models bring out acceptable predictive accuracy and LSTM is the best model for forecasting emergency patient arrivals in the selected hospital. The MAPE of LSTM model is 11.31%.