loading page

High Resolution Forecasting of Sparse Spatial-Temporal Events with Gaussian Processes and Random Fourier Features: A Solution to the NIJ Real-Time Crime Forecasting Challenge
  • +1
  • pauperium,
  • michaelchirico4,
  • cloef,
  • flaxter
michaelchirico4
Author Profile

Abstract

This article describes Team Kernel Glitches' solution to the National Institute of Justice’s (NIJ) Real-Time Crime Forecasting Challenge \cite{nij_real-time_2017}. The goal of the NIJ Real Time Crime Forecasting Competition was to maximize two different crime hotspot scoring metrics for calls-for-service to the Portland Police Bureau (PPB) in Portland Oregon during the period from March 1, 2017 to May 31, 2017. Our solution to the challenge consists of a spatiotemporal forecasting model using Random Fourier Features (RFF) with auto-regressive kernel densities. Model parameters, including random fourier features, spatial and temporal length scales, KDE lags, KDE bandwidths, as well as cell shape, size, and rotation, were learned using Vowpal Wabbit (VW). Resulting predictions exceeded baseline KDE estimates by XXXXX. Performance improvement over baseline predictions were particularly large for sparse crimes over short forecasting horizons.