An outlier detection algorithm based on arbitrary shape clustering

Abstract

Outlier detection is an important branch in data mining field. It provides new methods for analyzing all kinds of massive, complex data with noise. In this paper, an outlier detection algorithm is presented by introducing the arbitrary shape clustering approach and discussing the concept of abnormal cluster. The algorithm firstly partitions the dataset into several clusters by proposed clustering approach. Outliers are then detected from the cluster set according to the abnormal cluster concept. Moreover, by introducing inter-cluster dissimilarity measure, the proposed algorithm gains a good performance on the mixed data. The experimental results on the real-life datasets show our approach outperform the existing methods on identifying meaningful and interesting outliers. © 2009 Springer.