Predicting emergency medical service call demand: A modern spatiotemporal machine learning approach

Abstract

The primary goal of emergency medical service (EMS) agencies is to effectively allocate the ambulances and personnel required to provide sufficient geographic coverage of a service area while minimizing response times to high-priority call requests. Given that the demand for ambulances is known to fluctuate spatially and temporally based on the time of day and day of the week, EMS practitioners depend on call volume forecasts to develop staffing and dynamic redeployment plans. In this study, a series of daily, hourly, and spatially distributed hourly call volume predictions are generated using a multi-layer perceptron (MLP) artificial neural network model following feature selection using an ensemble-based decision tree model. For spatially distributed predictions, K-Means clustering is applied to produce heterogeneous spatial clusters based on call location and associated call volume densities. The predictive performance of the MLP model is benchmarked against both a selection of traditional time-series forecasting techniques and a common industry method. Results show that MLP models outperform time-series and industry forecasting methods, specifically at finer levels of spatial granularity where the need for more accurate call volumes forecasts is more essential.