Keywords: criminal phenomena, data mining, self–organising map, SOM, variable evaluation, k–means clustering, decision trees, logistic discriminant analysis, nearest neighbour searching, USA, United States, visualisation, crime development, socio–economic development
Application of the self–organising map to visualisation of and exploration into historical development of criminal phenomena in the USA, 1960–2007
Underneath the prevalence of criminal phenomena, many variables can be used to describe the background data such as the historical development of crime against socio–economic development. With large amount of data and evolution of data processing, multi–dimensional analysis becomes possible. Based on longitudinal (1960–2007), crime and socio–economic data (22 variables), we used the self–organising map (SOM) for development of criminal phenomena in the USA. Classification power of variables was evaluated and, e.g., k–means clustering was used for obtaining comparable results. After initial processing of the data with the SOM, six clusters of years were identified. We show how the SOM is applied to analysing criminal phenomena over a span of several decades. Results proved that, after the evaluation of variables for classification, and validation with k–means clustering, nearest neighbour searching, decision trees, and logistic discriminant analysis, SOM can be a new tool for mapping criminal phenomena processing multivariate data.