A comparative analysis of selected clustering algorithms for criminal profiling

Abstract

Several criminal profiling systems have been developed to assist the Law Enforcement Agencies in solving crimes but the techniques employed in most of the systems lack the ability to cluster criminal based on their behavioral characteristics. This paper reviewed different clustering techniques used in criminal profiling and then selects one fuzzy clustering algorithm (Expectation Maximization) and two hard clustering algorithm (K-means and Hierarchical). The selected algorithms were then developed and tested on real life data to produce "profiles" of criminal activity and behavior of criminals. The algorithms were implemented using WEKA software package. The performance of the algorithms was evaluated using cluster accuracy and time complexity. The results show that Expectation Maximization algorithm gave a 90.5% clusters accuracy in 8.5s, while K-Means had 62.6% in 0.09s and Hierarchical with 51.9% in 0.11s. In conclusion, soft clustering algorithm performs better than hard clustering algorithm in analyzing criminal data.
 Keywords: Clustering Algorithm, Profiling, Crime, Membership value