Rapid Associative Interactive Learner

Abstract

In the real world as we know that there is no single best automated classifier for all the datasets, instead there is this evident challenge of building intelligent interactive classifiers which can effectively work on a wide array of datasets. In the past, the KDD process has been totally focused on developing high accuracy classifiers neglecting the human involvement needed in building such systems. Associative classifiers possess two major advantages which are their high accuracy rates and the ease with which their rule based logic can be interpreted by human experts. Our paper induces user interaction into these learners and combines them in building a robust interactive associative classifier. We present the notion of two different kinds of class association rules (CAR) in interactive domains which are Soft and Hard CAR. We simultaneously study the impact of the involvement of these kinds of CAR's dynamically during the classification. We compare this approach to one of our previous endeavors SNAIL. We study the patterns of user pings in both the cases and conclude by stating how the current version in comparison to the previous strictly abides by the norms of an ideal interactive learner.