Evaluation of Artificial Intelligence Models and Classical Statistics Models of Time Series in Forecasting the Number of Hospital Inpatient Admissions

The study and analysis of each health system has become a necessity for its performance improvement through time. In this context, management and analysis of the number of patients is an important factor in the process of improving manager's decisions. The aims of this study were to explore and evaluate the use of multiple time series forecasting methods to predict monthly hospital inpatient admissions at six public hospitals in Mashhad city and to compare the accuracy performance of these methods.
This cross-sectional modeling study was performed based on monthly data of inpatient admissions at six public hospitals in Mashhad from March 2004 through March 2016. Data were extracted from database of the Statistics Office of Mashhad University of Medical Sciences. Holt-winters, Seasonal Autoregressive Integrated Moving Average (SARIMA), Multilayer Perceptron (MLP) and Generalized Regression Neural Networks (GRNN) models were applied to forecast monthly inpatient numbers at each hospital. The error of the models in regard to the predicted values was reported through Mean Absolute Percentage Error (MAPE).
Holt-Winters method, due to providing the optimal forecasting performance in four hospitals, could be an efficient method for predicting the number of inpatients in hospitals. Totally, the studied models with a MAPE from 2.13% to 4.12% showed acceptable performance in all six hospitals.
Time series analysis is an adequate practical tool for predicting the number of hospital inpatient admissions. Given the unique characteristics of different hospitals, applied methods in this study, including modeling and data analysis can be used in other hospitals to improve their resource allocation and strategic planning.