Finding Unexpected Patterns in Data
Balaji Padmanabhan, Alexander Tuzhilin · Studies in fuzziness and soft computing · 2002
Many pattern discovery methods in the KDD literature have the drawbacks of (1) discovering too many obvious or irrelevant patterns and (2) not using prior knowledge systematically. In this chapter we present an approach that addresses these drawbacks. In particular we present an approach to characterizing the unexpectedness of patterns based on prior background knowledge in the form of beliefs. Based on this characterization of unexpectedness we present an algorithm, ZoomUR, for discovering unexpected patterns in data.