Paper Title

Abstract - Forecasting demand for EMS, specifically ambulance response, is one of the most crucial aspects of prehospital care. There are random elements as well as many contextual factors, weather, demographics, special events, etc., that complicate forecasting. To have real-time deployment value, forecasts must predict at a fine level of granularity with respect to the temporal and spatial location. Emergency Medical Services managers have limited resources to accomplish defined goals such as ambulance response for emergencies to fall within defined times. Number and location of ambulances is one way decision-makers can efficiently meet response times. Call data from a metropolitan area in the southern United States with a time stamp and global positioning system (GPS) location field will be cleaned and organized to be fed into multiple types of Artificial Neural Network (ANN). These will be constructed and run on IBM SPSS Statistics v29©. Two different architectures will be compared with three different activation functions. The two architectures will be the feed-forward neural network (FFNN) and the recurrent neural network (RNN). The different activation functions will be hyperbolic tangent, sigmoid/logistic, and rectified linear unit (ReLU). Mean squared error (MSE) and mean absolute percent error (MAPE) will be used to compare the forecast accuracies. If the results indicate that the ReLU activation function is the most accurate, then a comparison will be performed among the different variations of ReLU, Leaky ReLU, Parametric ReLU, and Exponential Linear Units (ELU).Once the analysis has been completed the forecast outputs from the most accurate model will be the inputs for different static and dynamic ambulance deployment models to determine if there was an appreciable increase in ambulance coverage or minimization of resource use. Keywords - Emergency Medical Services, EMS calls, Artificial Neural Networks, Forecasting.