Dividing Outliers into Valuable and Noise Points

Abstract

A great number of different clustering algorithms exists in computer science. These algorithms solve the task of dividing data set into clusters. Data points which were not included into one of these clusters are called ‘outliers’. But such data points can be used for the discovery of unusual behavior of the analyzed systems. In this article we present a novel fuzzy based optimization approach for division these outliers into two classes: interesting (usable for solving the problem) outliers and noise.