Gaining insight in domestic violence with Emergent Self Organizing Maps

Abstract

Topographic maps are an appealing exploratory instrument for discovering new knowledge from databases. During the past years, new types of Self Organizing Maps (SOM) were introduced in the literature, including the recent Emergent SOM. The ESOM tool is used here to analyze a large set of police reports describing a wide range of violent incidents that occurred during the year 2007 in the Amsterdam-Amstelland police region (the Netherlands). This article aims to demonstrate that the ESOM tool provides a valuable exploratory instrument for examining unstructured text in police reports. First, it is shown how ESOM was used to discover a range of new features that better distinguish domestic from non-domestic violence cases. Second, it is demonstrated how this resulted in a significant improvement in classification accuracy. Third, the ESOM tool facilitates an in-depth investigation of the nature and scope of domestic violence, which is particularly useful for the domain expert. Interestingly, it was discovered that the definition of domestic violence employed by the management was much broader than the definition employed by police officers. Fourth, the ESOM tool enables an accurate and automated assignment of either a domestic or a non-domestic violence label to unclassified cases. Finally, ESOM is a highly accurate and comprehensible case triage model for detecting and retrieving wrongly classified cases.