Comparing the Forecasting Performance of Seasonal Arima and Holt -Winters Methods of Births at a Tertiary Healthcare Facility in Ghana

Abstract

Introduction: Studies have shown periodic variations in the number of births using different mathematical models. A study conducted at the Korle-Bu teaching hospital obtained Seasonal Autoregressive Integrated Moving Average (SARIMA) model on a monthly number of birth for an 11-year data. However, this study did not compare the obtained model with other forecasting methods to determine the method that will best explain the data. This study sought to compare seasonal SARIMA model with Holt-Winters seasonal forecasting methods for an 11-year time series data on the number of births.. Methods: Data were analysed in R software (version 3.3.3). Holt-Winters and seasonal ARIMA forecasting methods were applied to the birth data. The errors of the out – of-sample forecast of these methods were compared and the one with the least error was considered the best forecasting method. Results The in-sample forecasting errors showed that SARIMA (2,1,1) x (1,01,) was the best among the other models. The out-of-sample errors also showed that all the SARIMA models had lower errors compared to the Holt-Winters form of additive and multiplicative methods based on the forecasting accuracy indices of the monthly number of births for an 11-year period. It was also found that the months with very high statistically significant number of births over the period was from March to August. Conclusion: The SARIMA models were superior to the Holt-Winters models. This is essential for optimal forecasting of the number of births for planning and effective delivery of Obstetrics services.