Inductive Learning of Characteristic Concept Descriptions

This paper deals with the problem of learning characteristic concept descriptions from examples and describes a new generalization approach implemented in the system Cola-2. The approach tries to take advantage of the information which can be induced from descriptions of unclassiied objects using a conceptual clustering algorithm. Experimental results in various real-world domains strongly support the hypothesis that the new approach delivers more correct (and possibly more comprehesible) concept descriptions than exisiting methods, if the induced concept descriptions are also used to classify objects which belong to concepts which were not present in the training data set. This paper describes the generalization approach implemented in Cola and presents experimental results obtained with a relational and a propositional real world data set.