Class separation through variance: a new application of outlier detection

摘要: This paper introduces a new outlier detection approach and discusses extends concept, class separation through variance. We show that even for balanced concentric classes differing only in variance, accumulating information about the outlierness of points multiple subspaces leads to ranking which naturally tend separate. Exploiting this highly effective efficient unsupervised approach. Unlike typical algorithms, method can be applied beyond ‘rare classes’ case with great success. The algorithm FASTOUT number novel features. It employs sampling is efficient. handles arbitrarily sized converges an optimal subspace size use objective function. In addition, two approaches are presented automatically deriving data from ranking. Experiments typically outperforms other state-of-the-art methods on high-dimensional such as Feature Bagging, SOE1, LOF, ORCA Robust Mahalanobis Distance, competes leading supervised classification separating classes.