Fine-Grained Crime Prediction using Massive Ubiquitous Data Sources

• In Event: Big Data and Crime Prediction

Abstract

Current approaches in predictive policing, like PredPol or HunchLab, make use of historical crime data, socio-demographic data, as well as static geographical and temporal factors (like points of interest or day of the week). In this work, we investigate the feasibility of massive spatio-temporal data that make for better proxies of human activity in cities (like 311 calls or check-ins in local-based social networks). Informed by theories in criminology, we leverage these data sources to craft an extensive set of features and employ state-of-the-art machine learning techniques to predict crime in New York City. Crime is investigated at a fine-grained level, with data being analyzed at a census track and week granularity, and across several crime categories.

Authors

• Cristina Kadar, ETH Zurich, Switzerland

• Raquel Rosés, ETH Zürich, Switzerland

• Irena Pletikosa, ETH Zurich, Switzerland