Ryan Bissell-Siders (GREYC)

We describe an application of pattern-extraction to machine learning. Already applied to machine learning are: – mining for patterns which capture correlated sets of attributes, and thus describe all correlations in a dataset and – mining to discover concise explanations for classification. Mining for patterns which satisfy statistical properties in relation with other patterns has been used to describe trends in datasets or the differences between datasets, but have not been applied to machine learning. We have performed experiments in which a statistical test on the supports of subpatterns of a pattern are used to guide the classification. We will describe some context from machine learning, concept lattices, and contrast mining. We will describe our experiment. Time permitting, we will describe the intended application in chemoinformatics.