Forecasting Volume of Patients’ Visits in the Emergency Medicine Department at Muhimbili National Hospital, Tanzania

Abstract Background Emergency department overcrowding has developed as a severe and mounting problem common worldwide with substantial worldwide public health problems since it impedes the realization of Sustainable Development Goals of good health and well-being, including smooth planning and provision of quality services. Objective The main goal of this study is to forecast monthly patients’ visits in the Emergency Medicine Department at Muhimbili National Hospital, Tanzania to provide timely information to aid the decision-making process. Methodology Retrospective data of monthly patient visits registered in the Emergency Department at Muhimbili National Hospital from January 2016 to December 2020, with 60 observations, were collected. Time series methods of ARIMA and Multi-layer Perceptron (MLP) neural network models were used to analyze the data. The best model was chosen, and identified based on predicted accuracy between ARIMA and MLP models with the mean absolute percentage error (MAPE), root mean square error (RMSE), and mean absolute error (MAE). The software packages R Studio and Excel 2016 were used for statistical analyses. Results It was found that the MLP model with structure 2-(5)-1 performed better in the testing set (RMSE = 372.37, MAPE = 5.7282 and MAE = 304.70) compared to ARIMA (1,1,4) (RMSE = 426.69, MAPE = 6.9489 and MAE = 337.58 respectively). Also, the validation set of the MLP model with structure 2-(5)-1 performed better (RMSE = 763.89, MAPE = 17.3553 and MAE = 751.935) than ARIMA (1,1,4) (RMSE = 974.573, MAPE = 19.4181 and MAE = 893.9388 respectively). Conclusions Therefore, the MLP model with structure 2-(5)-1 was chosen for forecasting purposes whereby we predicted forecasts for 20 months. Finally, we recommend That the results of this study be adopted to help decision-makers and planners to be prepared and avoid dire situations.