IMPACT: A Novel Clustering Algorithm based on Attraction

Abstract

Clustering is a discovery process that groups data objects into clusters such that the intracluster similarity is maximized and the intercluster similarity is minimized. This paper proposes a novel-clustering algorithm, IMPACT (Iteratively Moving Points based on Attraction to ClusTer data), that partitions data objects by moving them closer according to their attractive forces. These movements increase separation among clusters while retaining the global structure of the data. Our algorithm does not require a priori specification of the number of clusters or other parameters to identify the underlying clustering structure. Experimental results show improvements over other clustering algorithms for datasets containing different cluster shapes, densities, sizes, and noise

D-IMPACT: A Data Preprocessing Algorithm to Improve the Performance of Clustering

“…IMPACT [14] is a two phases clustering algorithm which is based on the idea of gradually moving all data points closer to similar data points according to the attraction between them until the dataset becomes self-partitioned. In the first phase of the IMPACT algorithm, the data are normalized and denoised.…”