Search

Predicting demand in emergency medical services is crucial for saving people's lives. Most studies aggregate demand prediction within a zone, failing to offer insights at a more detailed level. This study aspires to fill this gap by introducing a novel, three-level, spatial-based approach that identifies the geographical location of expected emergency events. First, the proposed methodology introduces new concepts and notions to model emergency events, as sets of interconnected points in space, that create paths over time. Second, based on these paths, an artificial neural network, optimized using a new evolutionary algorithm, predicts the location of future demand (emergencies). Third, based on the predicted demand, a location-allocation model is applied to site ambulances prior to actual emergencies occurrence, enhancing thus location planning and decision making. This method is applied to a dataset comprising 2851 emergency events in Athens, Greece, and the outcomes are evaluated based on the actual emergency events occurred. Results show that the mean distance, between an actual emergency event and the nearest ambulance, located based on the expected demand as estimated by our approach, deviates by 110 m relative to the optimal solution. This deviation, adds only a few seconds of delay to the response time of an ambulance relative to the theoretically optimal solution (post hoc location). In addition, it improves the current solution (in which ambulances are waiting in a set of fixed location throughout the year), by >1 km, decreasing significantly response time. From a policy perspective, these results indicate that assessing expected emergency events through the proposed method, would allow medical services to optimally locate ambulances in advance, reducing response time and thus increasing survival rates and public safety. Unlabelled Image • Identifying the location (coordinates) of expected emergency events trough ANN. • ANN optimized by a new evolutionary algorithm. • 68.9% of the predicted emergency events lie closer than 150 m from the actual events. • Improved mean distance between an event and the nearest ambulance by 1 km. • Improved objective function cost by 19.7% (p-center model). [ABSTRACT FROM AUTHOR]