DBSCAN for Data with Tolerance

Abstract

In clustering, generally, we regard data in real space as one point in pattern space. Considering data uncertainty, it should be represented by a set instead of one point. Therefore, tolerance concept which means the permissible range of data is proposed by Endo et al[1, 2, 3]. By the way, DBSCAN[4], one of the hierarchical clustering algorithms, is one of the most widely used algorithms today. However, it can’t handle with data with tolerance. In this paper, we propose new clustering algorithm by introducing tolerance concept to DBSCAN and verify its effectiveness through numerical examples.