Mining 911 Calls in New York City: Temporal Patterns, Detection, and Forecasting

The New York Police Department (NYPD) is tasked with responding to a wide range of incidents that are reported through the city's 911 emergency hotline. Currently, response resources are distributed within police precincts on the basis of high-level summary statistics and expert reasoning. In this paper, we describe our first steps towards a better understanding of 911 call activity: temporal behavioral clustering, predictive models of call activity, and anomalous event detection. In practice, the proposed techniques provide decision makers granular information on resource allocation needs across precincts and are important components of an overall data-driven resource allocation policy.