Minimal Jumping Emerging Patterns: Computation and Practical Assessment

Bamba Kane (1) , Bertrand Cuissart (1) , Bruno Crémilleux (1)

1 Equipe CODAG - Laboratoire GREYC - UMR6072

Jumping Emerging Patterns (JEP) are patterns that only occur in objects of a single class, a minimal JEP is a JEP where none of its proper subsets is a JEP. In this paper, an efficient method to mine the whole set of the minimal JEPs is detailed and fully proven. Moreover, our method has a larger scope since it is able to compute the essential JEPs and the top-k minimal JEPs. We also extract minimal JEPs where the absence of attributes is stated, and we show that this leads to the discovery of new valuable pieces of information. A performance study is reported to evaluate our approach and the practical efficiency of minimal JEPs in the design of rules to express correlations is shown.

Mots clés: pattern mining; emerging patterns; minimal jumping emerg- ing patterns; ruled-based classification

Domaines: Base de données [cs.DB]

Bamba Kane, Bertrand Cuissart, Bruno Crémilleux. Minimal Jumping Emerging Patterns: Computation and Practical Assessment. 19th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2015), 2015, Ho Chi Minh City, Vietnam. ⟨10.1007/978-3-319-18038-0_56⟩. ⟨hal-01627079⟩