OutRules: A Framework for Outlier Descriptions in Multiple Context Spaces
Proceedings of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2012), Bristol, UK
Analyzing exceptional objects is an important mining task. It includes the identification of outliers but also the description of outlier properties in contrast to regular objects. However, existing detection approaches miss to provide important descriptions that allow human understanding of outlier reasons. In this work we present OutRules, a framework for outlier descriptions that enable an easy understanding of multiple outlier reasons in different contexts. We introduce outlier rules as a novel outlier description model. A rule illustrates the deviation of an outlier in contrast to its context that is considered to be normal. Our framework highlights the practical use of outlier rules and provides the basis for future development of outlier description models.