Short-Term Adaptive Emergency Call Volume Prediction

Sudden periods of extreme and persistent changes in the distribution of medical emergencies can trigger resource planning inefficiencies for Emergency Medical Services, causing delayed responses and increased waiting times. Predicting such changes and reacting adaptively can alleviate these adversarial impacts. In this paper, we propose a simple framework to enhance historically calibrated call volume models, the latter a focus of study in the arrival estimation literature, to give more accurate short-term prediction by refitting their residuals into time series. We discuss some justification of our framework from the perspective of doubly stochastic Poisson processes. We illustrate our methodology in predicting the hourly call volume to the 911 call center during the Covid-19 pandemic in NYC, showing how it could improve the performance of baseline historical estimators by close to 50% measured by the out-of-sample prediction error for the next hour.