Section: New Results
Mining Relevant Interval Rules
Participants : Philippe Besnard, Thomas Guyet, Véronique Masson, René Quiniou.
Rule mining is a classical data mining task. Numerical rule mining consists of extracting decision rules from a dataset with numerical attributes. In this work, we are interested in extracting a subset of accurate rules, called relevant rules. This selection criteria was introduced by Garriga et al. for categorical attributes . In  we extend the method of Garriga et al. for mining relevant rules on numerical attributes by extracting interval-based pattern rules. We proposed an algorithm that extracts such rules from numerical datasets using the interval-pattern approach from Kaytoue et al. . The algorithm has been implemented and intensively evaluated on real datasets. This study on numerical rules mining leads us to initiate a study about admissible generatizations of examples as rules .