Closest neighbors excluded outlier detection

Traditional distance-based outlier detection usually consider the distances from a point to its nearest neighbors as its outlier degree. In this case, if a few points form a small but dense cluster, which is far from other points, points in this small and dense cluster are not likely to be detected as outliers. In this paper, we propose a new distance-based outlier definition, Closest Neighbors Excluded (CNE) outlier, and the corresponding detection algorithm, which is able to detect dense outliers, as well as sparse outliers. Experimental results show that the CNE algorithm achieves great improvement in accuracy with little cost of efficiency.