An Efficient Clustering and Distance Based Approach for Outlier Detection

Outlier detection is a substantial research problem in the domain of data mining that aims to uncover objects which exhibit significantly different, exceptional and inconsistent from rest of the data. Outlier detection has been widely researched and finds use within various application domains including tax fraud detection, network robustness analysis, network intrusion and medical diagnosis. In this paper we propose an efficient clustering and distance based outlier detection technique. The clustering algorithms employed for this task are PAM, CLARA and CLARANS and a novel clustering algorithm I-CLARANS is proposed. The process of outlier detection is divided into two stages. In the first stage clustering is performed and in the second stage outlier detection is performed. The purpose is to perform clustering and outlier mining simultaneously. The experimental results depict that the proposed method is effective and promising in practice. We also present comparison of proposed algorithm with existing algorithms to validate its advantage in outlier detection. Keywords— Outlier detection, Data Mining, Clustering, PAM, CLARA, CLARANS.