Outlier detection in categorical data

Abstract

This chapter delves on a specific research issue connected with outlier detection problem, namely type of data attributes. More specifically, the case of analyzing data described using categorical attributes/features is presented here. It is known that the performance of a detection algorithm directly depends on the way outliers are perceived. Typically, categorical data are processed by considering the occurrence frequencies of various attributes values. Accordingly, the objective here is to characterize the deviating nature of data objects with respect to individual attributes as well as in the joint distribution of two or more attributes. This can be achieved by defining the measure of deviation in terms of the attribute value frequencies. Also, cluster analysis provides valuable insights on the inherent grouping structure of the data that helps in identifying the deviating objects. Based on this understanding, this chapter presents algorithms developed for detection of outliers in categorical data.