The traditional categorization of crime types relies on a hierarchical structure, from high-level categories to lower-level subtypes. This tree-based classification treats crime types as mutually independent when they do not branch from the same higher-level category, therefore lacking inter-category semantic relations. The issue then extends over crime distribution analysis of urban regions, often reporting statistics based on crime type counts, but neglecting implicit relations between different crime categories. Our study aims to fill this information gap, providing a more complete understanding of urban crime in both qualitative and quantitative terms. Specifically, we propose a vector-based crime type representation, constructed via unsupervised machine learning on temporal and geographic factors. The general idea is to define crime types as “related” if they often occur in the same area at the same time span, regardless of any initial hierarchical categorization. This opens to a new metric of comparison that goes beyond pre-defined structures, revealing hidden relationships between crime types by generating a vector space in a completely data-driven manner. Crime types are represented as points in this space, and their relative distances disclose stronger or weaker semantic relations. A direct application on urban crime distribution analysis stands out in the form of visualization tools for intuitive data investigations and convenient comparison measures on composite vectors of urban regions. Meaningful insights on crime type distributions and a better understanding of urban crime characteristics determine a valuable asset to urban management and development.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited