ON FUZZY NEIGHBORHOOD BASED CLUSTERING ALGORITHM WITH LOW COMPLEXITY

Abstract

The main purpose of this paper is to achieve improvement in the<br />speed of Fuzzy Joint Points (FJP) algorithm. Since FJP approach is a basis<br />for fuzzy neighborhood based clustering algorithms such as Noise-Robust FJP<br />(NRFJP) and Fuzzy Neighborhood DBSCAN (FN-DBSCAN), improving FJP<br />algorithm would an important achievement in terms of these FJP-based meth-<br />ods. Although FJP has many advantages such as robustness, auto detection<br />of the optimal number of clusters by using cluster validity, independency from<br />scale, etc., it is a little bit slow. In order to eliminate this disadvantage, by im-<br />proving the FJP algorithm, we propose a novel Modi ed FJP algorithm, which<br />theoretically runs approximately n= log2 n times faster and which is less com-<br />plex than the FJP algorithm. We evaluated the performance of the Modi ed<br />FJP algorithm both analytically and experimentally.