Co-training Based Attribute Reduction for Partially Labeled Data

Rough set theory is an effective supervised learning model for labeled data. However, it is often the case that practical problems involve both labeled and unlabeled data. In this paper, the problem of attribute reduction for partially labeled data is studied. A novel semi-supervised attribute reduction algorithm is proposed, based on co-training which capitalizes on the unlabeled data to improve the quality of attribute reducts from few labeled data. It gets two diverse reducts of the labeled data, employs them to train its base classifiers, then co-trains the two base classifiers iteratively. In every round, the base classifiers learn from each other on the unlabeled data and enlarge the labeled data, so better quality reducts could be computed from the enlarged labeled data and employed to construct base classifiers of higher performance. The experimental results with UCI data sets show that the proposed algorithm can improves the quality of reduct.