Computational Criminology

In this report, we present our work done in spring 2011 on the UK crimes dataset. This dataset was first released in December 2010, and contains reports of crimes committed in England and Wales, with their type and location. We first perform some exploratory analysis on this data, by looking at the correlation of crime rates with some independent variables, such as the population density or the unemployment rate, as well as the relationship between different types of crimes. We also study the spatial autocorrelation of the crime rates. Then, we define a classification problem in which we are interested in identifying probable criminals from mobility traces and aggregated crimes reports. We first introduce a basic algorithm to try and solve this problem, and then reformulate our model to fit a probabilistic group testing setup.