Towards Designing a Graph-Based Clustering of Crime Reports Using Graph Learning Techniques

Analyzing crime reports is critical for law enforcement to investigate cases and keep communities safe. However, sorting through thousands of reports manually is an extremely time-consuming and difficult task for analysts. While existing machine learning techniques can automatically categorize crime reports, most approaches focus exclusively on classification rather than finding deeper patterns by clustering similar reports together. Grouping reports based on contextual similarities provides investigators with valuable insights and improves how related information is retrieved and connected. Traditional clustering methods struggle when dealing with complex relationships and sparse data structures typical of crime datasets. This limitation has led researchers to explore graph-based clustering as a promising approach. By modelling the data as interconnected nodes in a graph, these methods can potentially uncover deeper insights. This study developed a comprehensive framework that includes collecting crime data, preprocessing, converting reports into data vectors, constructing graph representations, applying advanced graph learning algorithms, and finally clustering similar crime reports together. The proposed graph was evaluated to assess its effectiveness by using the clustering coefficient and density score.