An Experimental Approach for Evaluating the Performance of Clustering Techniques for Crime Detection

In the 21st century, when technology connects the entire world, our society suffers from modern age crime and criminals. A big challenge with the law enforcement agencies is to find and detect crimes. In machine learning, clustering algorithm is used to analyze crime related textual data focusing on identifying crime patterns from crime text data. In this research three clustering methods such as K-means, Agglomerative and DBSCAN are applied to crime reports to group similar description into clusters. These clusters were then analyzed to reveal the underlying patterns such as common crime type, crime time and crime location. The performance of these clustering techniques was evaluated using clustering assessing metrics Silhouette Score, ARI, homogeneity and completeness. It was observed that Silhouette score of K-Means, Agglomerative and DBSCAN were recorded 97%, 2.89% and 0.02%. The Adjust Rand Index of the three was 100%, 0% and 32.65%.The Homogeneity and completeness were 100%, 53.88%, 0.0% and 100%, 50.73%, 100%. The scores shows that K-means gives the better results in all the four metrics as compare to agglomerative and DBSCAN algorithm. Keywords: K-means, DBSCAN, Agglomerative, clustering, machine learning.