Univariate Time Series Analysis of Short-Term Forecasting Horizons Using Artificial Neural Networks: The Case of Public Ambulance Emergency Preparedness

This study examined the applicability of artificial neural network models in modelling univariate time series ambulance demand for short-term forecasting horizons in Zimbabwe. Bulawayo City Councils’ ambulance services department was used as a case study. Two models, feed-forward neural network (FFNN) and seasonal autoregressive integrated moving average, (SARIMA) were developed using monthly historical data from 2010 to 2017 and compared against observed data for 2018. The mean absolute error (MAE), root mean square error (RMSE), and paired sample - test were used as performance measures. Calculated performance measures for FFNN were MAE (94 . 0), RMSE (137 . 19), and the test statistic value (>0 . 05) whilst corresponding values for SARIMA were 105 . 71, 125 . 28, and ( < 0 . 05), respectively. Findings of this study suggest that the FFNN model is inclined to value estimation whilst the SARIMA model is directional with a linear pattern over time. Based on the performance measures, the parsimonious FFNN model was selected to predict short-term annual ambulance demand. Demand forecasts with FFNN for 2019 reflected the expected general trends in Bulawayo. The forecasts indicate high demand during the months of January, March, September, and December. Key ambulance logistic activities such as vehicle servicing, replenishment of essential equipment and drugs, staff training, leave days scheduling, and mock drills need to be planned for April, June, and July when low demand is anticipated. This deliberate planning strategy would avoid a dire situation whereby ambulances are available but without adequate staff, essential drugs, and equipment to respond to public emergency calls.