Compression-based measures for mining interesting rules

Abstract

An interestingness measure estimates the degree of interestingness of a discovered pattern and has been actively studied in the past two decades. Several pitfalls should be avoided in the study such as a use of many parameters and a lack of systematic evaluation in the presence of noise. Compression-based measures have advantages in this respect as they are typically parameter-free and robust to noise. In this paper, we present J-measure and a measure based on an extension of the Minimum Description Length Principle (MDLP) as compression-based measures for mining interesting rules. © 2009 Springer Berlin Heidelberg.