A Test Cost Sensitive Heuristic Attribute Reduction Algorithm for Partially Labeled Data

Attribute reduction is viewed as one of the most important topics in rough set theory and there have been many researches on this issue. In the real world, partially labeled data is universal and cost sensitivity should be taken into account under some circumstances. However, very few studies on attribute reduction for partially labeled data with test cost have been carried out. In this paper, based on mutual information, the significance of an attribute in partially labeled decision system with test cost is defined, and for labeled data, a heuristic attribute reduction algorithm TCSPR is proposed. Experimental results show the impact of test cost on reducts for partially labeled data and comparative experiments of classification accuracy indicate the effectiveness of the proposed method.