Crime Analysis Using Graph-Based Feature Selection

The current study extracts all potential crime attributes from reported offenses and selects just those that are needed for crime pattern assessment. To accomplish this, before using a crime report, it is processed, and criminal features are gathered to build feature vectors for each report. Based on a cosine relationship score, a weighted and undirected graph evolved, with the distinguishing qualities as vertices and the inverse of the similarity rating as the weight of the accompanying edge. The threshold number is determined by averaging all edge weights. Two separate subgraphs have been formed based on the threshold. The resulting two subgraphs could be a collection of several components that were used individually as input to the next iteration of the clustering operation. Each component’s threshold has been changed independently, and they have been further partitioned into more compact subgraphs. The Silhouette Index, a cluster validation metric, is computed at each level of iteration, and the technique occurs over and over just in case the cluster quality improves. Eventually, the graph clustering method categorizes features. Thus, the primary improvement of this paper is feature selection, which not only improves the efficiency but also the veracity of the following mining process.