Discernibility Matrix Approach to Exception Analysis

Exception plays an important role in Knowledge Discovery in Databases (KDD). As far as data description is concerned, exception could serve as the complement to rule to form concise representation about data set. More importantly, exception may provide more information than rule for people to understand the data set. This paper defines text as the description of data set and considers some special objects as exceptions to complement text. An approach based on the distribution of core in discernibility matrix of text is presented to identify exceptions, and the exceptions selected by this way do bear some interesting properties.