A supervised learning approach for criminal identification using similarity measures and K-Medoids clustering

Data mining is an efficient tool for extracting the features from large data sets. It is an efficient approach to discover the hidden relationship from the known data and construct new knowledge discovery. In the field of crime detection and criminal identification, data mining tools and algorithms are plays major role. Criminology is a process to identify the crime characteristics and behavior. We have large volume of crime datasets with many attributes leads to more complexity for crime character identification process. Identification of criminal is the main process for further improvement in the police investigation. Data mining tools are best approach for Criminal identification based on characteristic and nature of crime. In this paper, we have proposed a supervised approach for identifying the suspected list of criminal's using similarity measure and K-Medoids cluster algorithm. K-Medoids clustering algorithm groups the more closely related crimes as an individual group and each group will have unique set of features. The unique features set is used for identification of criminal using similarity measure algorithms based on distance measure. The proposed system has two phase, training and testing phase. In this approach, we have trained the proposed system with supervised data set with collected crime information from various places of Tamil Nadu through online available data. In the testing phase, first identify the cluster closest to the test crime by using K-Medoids clustering algorithm and then identify the suspected criminal list using similarity measure. The initial stage of implementation and analysis of the proposed scheme provides good results and high accuracy. The proposed scheme is compared with related K-Means clustering algorithm with same set of training and test.