Criminal pattern identification based on modified K-means clustering

Abstract

Data mining methods like clustering enable police to get a clearer picture of criminal identification and prediction. Clustering algorithms will help to extracts hidden patterns to identify groups and their similarities. In this paper, a modified k-mean algorithm is proposed. The data point has been allocated to its suitable class or cluster more remarkably. The Modified k-mean algorithm reduces the complex nature of the numerical computation, thereby retaining the effectiveness of applying the k-mean algorithm. Firstly, the data are extracted from the communications and movements record after tracking the park visitors over three days. Then, the original data will be visualised in a graphical format to help make a decision about how many numbers to consider as the K cluster. Secondly, the modified k-means algorithm on the clusters initial centre sensitivity will be performed. This will link similar segments and determine the occurrence of each data point in every segment group rather than partitioning the entire space into various segments and calculating the occurrence of the data point in every segment. Thirdly, result checking and a comparison with the normal k-mean will be performed. The investigation will focus on the movement of people around the park where the crime occurred, and how people move and communicate in the park, how patterns change, and the movement of groups and individuals. The experiments show that the modified K-means algorithm leads to a better way of observing the data to identify groups and their similarities and dissimilarities in the criminal dataset as a specific domain.