2013 Volume 78 Issue 686 Pages 957-967
We analyze the relationship between bag snatching and spatial relationships in Kyoto City. In particular, we model natural surveillance with various facade components. We then analyze the data as a classification problem that divides space into areas in which crimes occur and areas in which crimes do not occur. Since there exists fuzziness in the categorization of space, we propose a new approach that combines clustering and classification based on semi-supervised learning, which determines the categorization so as to improve classification accuracy. In addition, we use an emerging pattern based classifier and obtain spatial features related to crime.