Learning Classification Rules for Multiple Target Attributes

Abstract

Among predictive models, ‘if-then’ rule sets are one of the most expressive and human readable model representations. Most of the existing approaches for rule learning focus on predicting a single target attribute/class. In practice, however, we encounter many problems where the task is to predict not one, but several related target attributes. We employ the predictive clustering approach to learn rules for simultaneous prediction of multiple target attributes. We propose a new rule learning algorithm, which (unlike existing rule learning approaches) generalizes to multiple target prediction. We empirically evaluate the new method and show that rule sets for multiple target prediction yield comparable accuracy to the respective collection of single target rule sets. The size of the multiple target rule set, however, is much smaller than the total size of the collection of single target rule sets.