Abstract
In rough set approaches, decision rules are induced from a given data table showing the relation between attribute values and classes of objects. The induced decision rules are used for the classification of new objects by their attribute values. However, because the given data table does not always include all possible patterns, some of new objects do not match any decision rule conditions. In those cases, no estimated classes are obtained. Classes of such new objects are estimated by using partially matched decision rules. In this paper, to enhance the classification accuracy, we propose to add supplementary rules which can work well for the unmatched new objects in the class estimation. We define the supplementary rules and propose a method for inducing them. We examine the performance of the classifier with supplementary rules by comparisons with the classifier without supplementary rules.
