Pattern Mining in Numerical Data: Extracting Closed Patterns and their Generators
Mehdi Kaytoue, Sergei O. Kuznetsov, Amedeo Napoli · HAL (Le Centre pour la Communication Scientifique Directe) · 2010
In this paper we study the extraction of closed patterns associated to their generators in numerical data. Many works have addressed the problem of extracting itemsets for generating association rules. Considering numerical data, an appropriate discretization is most of the time necessary, in order to split attribute ranges into intervals maximizing some interest functions, e.g. support, confidence, or other statistical measures. We investigate here an alternative point of view using pattern structures in Formal Concept Analysis. Pattern structures can be efficiently used to extract closed patterns without any prior discretization. Two original and efficient algorithms for characterizing frequent closed patterns and their generators in numerical data are proposed and experimented. Finally, we conclude showing the usefulness of such patterns in classification problems and privacy preserving data-mining.