Attribute-Based K-Means Algorithm

Abstract

Clustering is a method to discover hidden natural structure in a dataset of a phenomenon. In this study, we have extended K-Means algorithm for spatiotemporal dataset by introducing attribute-based mass function to calculate center of mass of cluster instead of calculating geometry-based centroid in the dataset. The proposed modification in traditional K-Means algorithm produces more meaningful clusters and converges faster than traditional K-Means. In our study, we have used a real ‘fire dataset’ to conduct experiments on the proposed approach for clustering.