LDBOD: A novel local distribution based outlier detector

As an important research direction in KDD field, outlier detection has been drawing much attention from different communities. In this paper, two novel algorithms LDBOD and LDBOD+ for outlier detection are proposed. Similar to LOF, they also aim to find local outliers. However, LDBOD/LDBOD+ detects local outliers from the viewpoint of local distribution, which is characterized through three proposed measurements, local-average-distance, local-density, and local-asymmetry-degree. Several experiments were conducted to demonstrate the advantages of LDBOD/LDBOD+ compared with LOF.