Local governments have a responsibility to provide emergency medical services (EMS) to the citizens of their region. The major obstacle facing the men and women that administer EMS is the allocation of limited resources. They must determine staffing of emergency medical technician (EMT) personnel and placement of emergency vehicles to ensure a certain level of service. Service is commonly defined as response time to an emergency call. Much operations research (OR) has been done in the area of ambulance deployment to minimize response times. The models for ambulance location and deployment depend on accurate demand data to be effective. Not only must the forecast be accurate, but it must also be in meaningful levels of temporal and spatial aggregation.

For these deployment schemes to work, the models used for forecasting future EMS demand must be accurate. Accurate forecasts will allow EMS managers to assign resources that will balance cost vs. satisfactory service levels. EMS demand forecasting has not garnered much exploration in the past two decades. Forecasting models have mainly been limited to some form of regression; however, inherent limitations of regression present the opportunity for alternate methods. In this study we investigate the use of artificial neural network (ANN) designs to forecast demand volume at a level of granularity that will improve the actual utility of existing EMS deployment models and compare the results to existing practices for accuracy of prediction.