Risk discovery of car-related crimes from urban spatial attributes using emerging patterns

Using spatial data mining techniques, we analyze car-related crimes such as auto theft, auto parts theft, and breaking into a car in the area of Nishikyo-ku, Kyoto City. The strategy of natural surveillance proposed by Crime Prevention Through Environmental Design (CPTED) is taken into consideration as visibility attributes. From the viewpoint of risk discovery, we do not employ ordinary association rule but a new data mining technique called Emerging Patterns (EPs). EP is defined as an itemset whose support increases significantly from one dataset to another. Since a large number of EPs are generated in general as in association rule, it is difficult to identify the critical factors which affect crime occurrences. Therefore, we will introduce two new ideas; (a) appropriately aggregating several EPs with high growth-rate and (b) identifying a pair of similar patterns A and B such that A is not associated with high crime occurrences while B is highly correlated with crime occurrences. Finding such similar patterns reveals that the attribute value which is in B but not in A is then identified as a critical factor which arouses crimes when combined with certain factors.