A Clustering Algorithm for Multi-density Datasets

Abstract

AbstractDBSCAN algorithm discovers clusters of various shapes and sizes. But it fails to discover clusters of different density. This is due to its dependency on global value for Eps. This paper introduces an idea to deal with this problem. The offered method estimates local density for a point as the sum of distances to its k-nearest items, arranges items in ascending order according to their local density. The clustering process is started from the highest density point by adding un-clustered points that have similar density as first point in cluster. Also, the point is assigned to current cluster if the sum of distances to its Minpts-nearest neighbors is less than or equal to the density of first point (core point condition in DBSCAN). Experimental results display the efficiency of the proposed method in discovering varied density clusters from data.KeywordsClustering methodsData analysisData miningKnowledge discoveryUn-supervised learning