Rank-Based Outlier Detection

We propose a new approach for outlier detection, based on a new ranking measure that focuses on the question of whether a point is “important” for its nearest neighbors; using our notations low cumulative rank implies the point is central. For instance, a point centrally located in a cluster has relatively low cumulative sum of ranks because it is among the nearest neighbors of its own nearest neighbors. But a point at the periphery of a cluster has high cumulative sum of ranks because its nearest neighbors are closer to the points. Use of ranks eliminates the problem of density calculation in the neighborhood of the point and this improves performance. Our method performs better than several density-based methods, on some synthetic data sets as well as on some real data sets.