Mining Crime Data by Using New Similarity Measure
Guangzhu Yu, Shihuang Shao, Bing Jin Luo · 2008
Clustering technique was introduced to the field of crime data analysis for finding crime suspects, but compared with the high requirements for public security work, traditional clustering methods used in existing application systems provide a relatively low accuracy. To solve the problem, we proposed a new similarity measure method, i.e., segmented multiple metric similarity measure (SMMSM), to improve the accuracy of similarity measure. In our method, attributes are divided into different groups according to their importance to similarity, compensation relationships does not exist among attributes in different groups. The new measure is scalable with dimensionality of data and is both suitable for numeric data and categorical data. Experiments on real data show that our method has higher accuracy than other measures.