期刊名称:Current Journal of Applied Science and Technology
印刷版ISSN:2457-1024
出版年度:2017
卷号:18
期号:4
页码:1-8
语种:English
出版社:Sciencedomain International
摘要:Aims: Analysis of hospital processes has become a necessity to improve its performance over time through developing improved policies and decisions. In this context, the knowledge gained from an accurate prediction of patient flows would provide valuable information for strategic planning. This study aims at exploring and evaluating the predictability of time-series artificial neural network (ANN) approach for hospital inpatient flow forecasting using the partial autocorrelation function (PACF) of time-series data to identify the relevant time lags of the series as ANN inputs. Methodology: We collected retrospective data of the number of monthly inpatient flows from 2004 to 2015 of four hospitals. We evaluated the application of the ANN model that uses extracted PACFs from time series data to determine appropriate inputs for ANN. This approach was compared with the neural network auto-regression which automatically selects relevant lags. The performance of the ANN models was measured based on the mean absolute percentage error (MAPE) accuracy measure.Results: We used the ANN models for predicting monthly inpatient flows for first three months of 2016. The post sample analysis revealed that the ANN model using selected input variables based on the PACF analysis offered improvements in monthly inpatient flows predictions than neural network auto-regression. Totally, for all four hospitals, the integrated model of PACF-ANN had a MAPE ranging from 2.91% to 6.67%, indicating an accurate prediction. Conclusion: The ANN model with inputs extracted from the PACF analysis performs well for estimation of hospital inpatient flows. According to the unique characteristics of different hospitals, performance of the ANN model can vary from hospital to hospital. However, the proposed method of selecting input variables for the ANN model in this study may assist other hospitals and emergency departments for forecasting purposes.