Prediction of Crime Occurrence using Information Propagation Model and Gaussian Process

Abstract

Prediction and prevention of crime have long been one of the main concerns of public security and safety. Due to the emergence of available data and analytic tools, research for crime prediction has been attracting more attention recently. In essence, the current techniques are based on either the analysis of discrete crime event locations or the theory with aggregated crime data. However, it is difficult to estimate the probability of future crimes based on the direct interpretation of the past crime rate. Therefore, existing methods are not good at adapting to different environment and trends of crime occurrence. Currently, there is no standard method that can simultaneously address all challenges posed by different crime data sets. A more universal solution, which can cope with the changes in the environment and the diversity of crime occurrence would be highly desirable. In this paper, we present a novel approach to crime prediction and establishes a model flexible enough to apply to different circumstances. To achieve our goal, we build an information propagation model which incorporates a concept of information entropy. This research helps security organizations to address or react to crime occurrence proactively and helps local policy-makers to prevent or manage crime risks, which would eventually improve public security and safety.