Outlier detection using neighborhood radius based on fractal dimension

Partition outlier using neighborhood radius has proven to be an effective distance-based detection algorithm. However, it is not yet clear how to choose the neighborhood radius d_min, and getting the value by trial and error is still been widely adopted. This paper presents a method to get the neighborhood radius from fractal dimensions which is used to describe the self-similarity of a dataset. We first discuss how to calculate the fractal dimensions and how to value d_min, and then we use this value in distance-based outlier detection algorithms. Finally, we verify the validity of this neighborhood radius calculation method by experimental results.