Rule Induction through Clustering Classes for Nominal and Numerical Data

In this paper, we investigate the performance of rule induction based on a hierarchical structure of classes. Given a decision table including numerical and nominal attributes, a rule induction approach via clustering classes is proposed. By the employment of an agglomerative hierarchical clustering algorithm, a hierarchical structure of classes is extracted. MLEM2 which can accommodate numerical and nominal attributes is employed as a rule induction algorithm. Numerical experiments are executed in order to compare the proposed approach with a standard application of MLEMl and n²-classifier. Based on the experimental results and the construction of classifiers, characteristics of the proposed approach are described.