An auto-stopped hierarchical clustering algorithm integrating outlier detection algorithm

Abstract

It is a critical problem for the clustering analysis techniques to select the appropriate value of parameters. Meanwhile, the clustering algorithms lack the effective mechanism to detect outliers while treating outliers as "noise". By regarding outliers as valuable information, the paper proposes a novel hierarchical clustering algorithm that integrates a new outlier-mining method. The algorithm stops clustering according to the dissimilarity reflected by the detected outliers and needs only one parameter, whose appropriate value can be decided in the outlier mining process. After discussing some related topics, the paper adopts 5 real-life datasets to evaluate the performance of the clustering algorithm in outlier mining and clustering and compare it with other algorithms. © Springer-Verlag Berlin Heidelberg 2005.