A twostep clustering algorithm as applied to crime data of South Africa

Abstract

This study applied a TwoStep cluster analysis on the 29 serious crimes reported at 1119 police stations across South Africa for the 2009/2010 financial year. Due to this high number of variables and observations, it becomes difficult to apply some statistical methods without firstly using others as precursors. Classical methods have also been found to be inefficient as they do not have the ability to handle large datasets and mixture of variables. The AIC and BIC automatically identified the three clusters of crimes. The findings may guide authorities when developing interventions tailored to better meet the needs of individual cluster of crimes. Existing plans may also be enhanced to the advantage of residents. More emphasise may be placed on crimes that pose a serious threat. The SAPS may use these findings when reporting on national crime statistics. For future studies, discriminant analysis can be applied to check the clusters’ validity.