Abstract
Detection of outliers and anomalous behavior is a well-known problem in the data mining and statistics fields. Although the problem of identifying single outliers has been extensively studied in the literature , little or some effort has been devoted to the detection of small groups of outliers that are similar to each other but markedly different from the entire population. Many real world scenarios have small groups of outliers , e. g. a group of students that excel in a classroom or a group of spammers in an online social network. In this paper , we propose a novel method to solve this challenging problem that lies at the frontiers of outlier detection and clustering of similar groups. The method transforms a multidimensional dataset into a graph , applies a network metric to detect clusters and renders a representation for visual assessment to find rare events. We test the proposed method to detect pathologic cells (e. g. Cancer , HIV , CVA , etc .) in the biomedical science domain. The results are very promising and confirm the available ground truth provided by the domain experts.