No Detail Found!

Optimal initial centroid in k-means for crime topic.

Author(s)
Masnizah Mohd Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Malaysia
Qusay Walid Bsoul Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Malaysia
Shahrul Azman Mohd Noah Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Malaysia
Saidah Saad Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Malaysia
Nazlia Omar Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Malaysia
Mohd Juzaiddin Ab-Aziz Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Malaysia
Nazlena Mohamad Ali Institute of Visual Informatics, Universiti Kebangsaan Malaysia, Malaysia
Abstract
A wide number of different clustering method applications and their effectiveness in crime topics have been examined in this paper. Several works have investigated the optimal initial centroid of clustering crime topics. In this paper, we have compared the effectiveness of single pass clustering and k-means in detecting crime topics and aiding in the identification of events or crimes. We have also experimented on enhanced k-means clustering, in order to select the optimal initial centroid to be automatically compared with regular k-means, to choose the initial centroid randomly. Based on the main findings of this study, it was revealed that the experimental method, which was based on k-means, was proved to be better and more effective than single pass clustering in detecting and identifying events or crimes. For the initial number of centroids, it was found that the proposed method was more effective when used in selecting terms that were more than the number of topics, than when they were less. However, the best result was obtained when choosing a number of topics equal to the number of original topics. This implies that the optimal accuracy of clustering is achieved when selecting a large number of documents that have terms better than randomly chosen documents as a centroid.
Publication Details
Page(s) 19-26
DOI DOI not available
Published Journal: Journal of Theoretical and Applied Information Technology, Volume: 45, Issue: 1, Year: 2012

Request the full text from PASTIC

More from Journal of Theoretical and Applied Information Technology
0

Citations

0

Downloads

34

Views

Copyright ©  PASTIC National Center, Islamabad www.pastic.gov.pk