Outlier detection algorithm for categortical data using a granular computing theory

Many distance-based outlier detection algorithms were proposed in the past, however, these algorithms can not effectively deal with categorical data set. In this paper, we propose a novel formulation for the outlier degree of the objects that is based on the granular computing theory. We rank each object on the basis of its Outlier Factor and declare the top k objects in this ranking to be outliers. In order to developing relatively straightforward solutions to finding outliers from categorical data set, we construct an algorithm, named ODAGS (Outlier Detection Algorithm Based on Granular Set). Theory analysis and example calculation both manifest that the ODAGS is efficient and feasible.