Weight-based method for inside outlier detection

Outlier detection becomes more and more important in our real life, such as network intrusion detection and credit card fraud detection, etc. In this paper, a weight-based method is proposed for inside outlier detection. According to the concepts of density and volume information, the weight is defined and introduced to construct a new measure of outlier-ness. Firstly, the total weight of a given object p and its neighbors is computed via their volume and average density. Then the estimated weight of the neighbors is obtained via the neighborhood's volume and p's density. If the total weight is not close to the estimated weight, p is an outlier. The weight-based method shows more superiority in inside outlier detection than LOF in low dimensions. Moreover, the proposed method performs as well as LOD in a high-dimensional space or when no inside outlier exists.