This report proposes a generic spatiotemporal event crime forecasting method, which was submitted for the National Institute of Justice's (NIJ's) Real-Time Crime Forecasting Challenge of 2017.
The goal of the challenge was to predict several different types of crime hotspots over varying timeframes for Portland, OR, using calls-for-service data from the Portland Police Bureau (PPB). The proposed solution is a spatiotemporal forecasting model that combines scalable randomized Reproducing Kernel Hilbert Space (RKHS) methods for approximating Gaussian processes with autoregressive smoothing kernels in a regularized supervised learning framework. Although the smoothing kernels capture the two main approaches in current use in crime forecasting - heatmap-based (kernal density estimation and the self-exciting point process (SEPP) models - the RKHS component of the model can be viewed as an approximation to the popular log-Gaussian Cox Process model. For inference, the proposed method discretizes the spatiotemporal point pattern and learns a log intensity function using the Poisson likelihood and highly efficient gradient-based optimization methods. Model hyper-parameters were learned using temporal cross-validation, including quality of RKHS approximation, spatial and temporal kernel length-scales, number of autoregressive lags, bandwidths for smoothing kernels, and cell shape, size, and rotation. Resulting predictions significantly exceeded baseline kernel density estimation (KDE). This project had winning submissions to the competition in each of the different crime types, suggesting that the proposed method is generally applicable. Performance improvement over baseline predictions and competing submissions were particularly large for sparse crimes over short forecasting periods. 12 figures, 2 tables, and 92 references
Downloads
Similar Publications
- PROPERTY CRIME AND ECONOMIC BEHAVIOR - SOME EMPIRICAL RESULTS
- Violent Crime Reduction Roadmap: Working Together to Build Safer Communities
- Audit of the Office of Justice Programs’ Procurement Awarded to ICF Incorporated, LLC, to Support the Office for Victims of Crime Training and Technical Assistance Center