05547037

a

05547037 Suzuki, Einoshin Negative encoding length as a subjective interestingness measure for groups of rules. Theeramunkong, Thanaruk (ed.) et al., Advances in knowledge discovery and data mining. 13th Pacific-Asia conference, PAKDD 2009, Bangkok, Thailand, April 27--30, 2009. Proceedings. Berlin: Springer (ISBN 978-3-642-01306-5/pbk). Lecture Notes in Computer Science 5476. Lecture Notes in Artificial Intelligence, 220-231 (2009). 2009 Berlin: Springer EN

doi:10.1007/978-3-642-01307-2_22

Summary: We propose an interestingness measure for groups of classification rules which are mutually related based on the Minimum Description Length Principle. Unlike conventional methods, our interestingness measure is based on a theoretical background, has no parameter, is applicable to a group of any number of rules, and can exploit an initial hypothesis. We have integrated the interestingness measure with practical heuristic search and built a rule-group discovery method CLARDEM (Classification Rule Discovery method based on an Extended-Mdlp). Extensive experiments using both real and artificial data confirm that CLARDEM can discover the correct concept from a small noisy data set and an approximate initial concept with high ``discovery accuracy''.