DBSCAN-KNN-GA: a multi Density-Level Parameter-Free clustering algorithm

Abstract

Abstract DBSCAN is a popular tool to analyse datasets which can effectively discover clusters with arbitrary shapes. However, it requires two input parameters which are difficult to be determined, according to the fact that the performance of clustering result depends heavily on user-specified parameters. In addition, it uses global parameters which are not appropriate to those multi-density datasets. Aiming at these problems, we propose a parameter-free algorithm to perform DBSCAN with different density-level parameters. We select some classical datasets and a TLC taxi trip record used for experiments to compared our proposed algorithm with the original DBSCAN to evaluate the performance of our improved DBSCAN. The results show that the proposed algorithm is capable for efficiently and effectively detecting clusters automatically with variable density-levels. Compared with original DBSCAN, the proposed algorithm can discover more noise points and its execution accuracy is higher.