Spatially embedded co-offence prediction using supervised learning

摘要: Crime reduction and prevention strategies are essential to increase public safety reduce the crime costs society. Law enforcement agencies have long realized importance of analyzing co-offending networks---networks offenders who committed crimes together---for this purpose. Although network structure can contribute significantly co-offence prediction, research in area is very limited. Here we address important problem by proposing a framework for prediction using supervised learning. Considering available information about offenders, introduce social, geographic, geo-social similarity feature sets which used classifying potential negative positive pairs offenders. Similar other social networks, networks also suffer from highly skewed distribution pairs. To class imbalance problem, identify three types criminal cooperation opportunities help ratio significantly, while keeping half co-offences. The proposed evaluated on large dataset Province British Columbia, Canada. Our experimental evaluation four different show that novel features best predictors. Overall, experimentally high effectiveness framework. We believe our will not only allow law improve their strategies, but offers new criminological insights into link formation between