Applying Genetic Algorithm to Conceptual Clustering

Abstract

In this paper, we propose the use of Genetic Algorithm (GA) as a key element in the design of robust learning system. We describe a GA-based concept learner called GAIL, capable of inducing descriptive concepts in conjunctive statements. A motivation to use a GA is that GAs are able to explore large domains because they make use of a multipoint search strategy and are suitable to exploit massive parallelism. By incorporating a GA with a traditional method, the system has several important features. First, the system is robust. Second, the system introduces the modified evaluation criteria to evaluate the quality of a clustering. Finally, the system searches for an individual which refers to all the concepts at the same time. The effectiveness of this approach is illustrated by two exemplary problems and the clusters obtained with this system correspond well to categorization satisfied by people.