Induction of "In-between" Classes: Learning Vague Concepts

This paper presents a synergistic iterative process, SIR, for resolving between classes assigned to cases. The "vagueness" of concepts, represented by multi-class assignment to cases, is a common phenomenon in the context of concept learning from examples (CLFE) paradigm. The causes could be attributed to the specifics of the application domain, to the poor initial representation or, to the learning heuristics themselves. The methodology presented in this paper take advantage of multi-class assignment in order to improve learning results. Our methodology implements a two-step iterative process: (1) an inductive algorithm runs on the training set of cases, and (2) application of a specially devised set of heuristics aiming to invent new classes, and resolve the conflict presented by multi-class assignment. Thus, "vague" concepts, laying "in-between" the underlying concepts, are learned. Experiments on real-world domains from medicine and finance are presented, and the utility of the SIR process in decision-making tasks is discussed.