Mining of attribute interactions using information theoretic metrics

Knowledge of the statistical interactions between the attributes in a data set provides insight into the underlying structure of the data and explains the relationships (independence, synergy, redundancy) between the attributes. In a supervised learning problem, normally, a small subset of the classifying attributes are actually associated with the class label. Interaction information among the attributes captures the multivariate dependencies (synergy and redundancy) among the attributes and the class label. Mining the significant statistical interactions that contain information about the class label is a computationally challenging task - the number of possible interactions increases exponentially and most of these interactions contain redundant information when a number of correlated attributes are present. In this paper, we present a data mining method (named IM or Interaction Mining) to mine non-redundant attribute sets that have significant interactions with the class label. We further demonstrate that the mined statistical interactions are useful for improved feature selection as they successfully capture the multivariate inter-dependencies among the attributes.