Sequential three-way classifier with justifiable granularity

Abstract Sequential three-way decisions approach has been demonstrated as an effective methodology of human problem solving with the aid of multiple levels of granularity. Searching an appropriate information granularity for decision or classification is a crucial issue. In this paper, inspired by the principle of justifiable granularity, we investigate a novel classification algorithm, called sequential three-way classifier with justifiable subspace. The major contribution of this study is threefold. First, in training model, with an investigation of the essence of information granularity in rough sets theory, the justifiable attribute subspace is located in an interval with local and global notions. Second, in light of the advantages of attribute reduction technology, the local and global attribute subspaces are determined by core and reduct, respectively. Third, a novel dynamic tri-partition-based predicting strategy is presented with the aid of neighborhood monotonic property. Finally, several experiments are undertaken to verify the effectiveness of the proposed method. Compared with several state-of-the-art classifiers, the proposed algorithm generally exhibits a better classification performance involving fewer attributes.