Abstract
In classification problems on rough sets, the effectiveness of ensemble learning approaches such as bagging and random forests have been reported. We focus on occurrences of deficiencies in rows or columns of decision tables in these ensemble approaches. In this paper, we generalize these deficiencies to deficiencies of cells and propose an ensemble learning approach based on missing-valued decision tables. Decision rules are extracted by means of MLEM2 algorithm and two binary relations, that is, the similarity relation and the tolerance relation. We confirmed the effectiveness of the methods for classification performance through numerical experiments. Furthermore, we consider the robustness of these methods in absences of attribute values.
