Forecasting of patient arrivals at emergency department using firefly algorithm and artificial neural networks

Emergency services, which are usually the first place to consult in hospitals, are systems that are difficult to plan. Therefore, patient arrivals need to be predicted so that decision makers can effectively use personnel and materials in emergency departments. This study aims to estimate the number of patients applying to an emergency department. For this purpose, artificial neural networks models, which perform better than traditional methods in terms of prediction, have been used in recent years. On the other hand, the artificial neural network was trained with the firefly algorithm (FFA-ANN), which stands out in terms of training performance. The performance of this model is based on time series analysis methods such as autoregressive integrated moving average (ARIMA), Holt-Winters (HW) and exponential smoothing (EMA); It was compared with random forest (RF), multilayer perceptron (MLP), and long short-term memory (LSTM) methods, which are machine learning methods. To evaluate the prediction models, 1673 days of 'number of patients coming to the emergency department' data of a private hospital in Konya was used. The data set consists of data between 2018 and 2022. The data set is divided into two: the normal period between 2018-2019-2022 and the pandemic period between 2020-2021. The data set has been rearranged according to 22 disease groups determined by the World Health Organization. Pareto analysis was performed for 22 disease groups, and as a result of the analysis, 5 disease groups were selected to be used in prediction models. The models were created as time series prediction models consisting of a single input variable (the number of daily patients arriving to the emergency department in the 5 selected disease groups) and a single output variable. In order to determine the most successful prediction method, root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) performance criteria were used. As a result of the analysis, it was seen that FFA-ANN method for normal period data gave more effective results than other methods in predicting the number of admissions to the emergency department. Finally, a decision support system that makes predictions with the FFA-ANN model was constructed and an application was designed where the prediction output can be viewed by decision makers.