Fuzzy Core DBScan Clustering Algorithm

Abstract

In this work we propose an extension of the DBSCAN algorithm to generate clusters with fuzzy density characteristics. The original version of DBSCAN requires two parameters (minPts and ε) to determine if a point lies in a dense area or not. Merging different dense areas results into clusters that fit the underlined dataset densities. In this approach, a single density threshold is employed for all the datasets of points while the distinct or the same set of points can exhibit different densities. In order to deal with this issue, we propose Approx Fuzzy Core DBSCAN that applies a soft constraint to model different densities, thus relaxing the rigid assumption used in the original algorithm. The proposal is compared with the classic DBSCAN. Some results are discussed on synthetic data. © Springer International Publishing Switzerland 2014.