DualPOS: A Semi-supervised Attribute Selection Approach for Symbolic Data Based on Rough Set Theory

摘要

Rough set theory, supplying an effective model for representation of uncertain knowledge, has been widely used in knowledge engineering and data mining. Especially, rough set theory has been used as an attribute selection method with much success. However, current rough set approaches for attribute reduction are unsuitable for semi-supervised learning as no enough labeled data can guarantee to calculate the dependency degree. We propose a new attribute selection strategy based on rough sets, called DualPOS. It provides mutual function mechanism of multi-attributes, and generates the most consistent one as a candidate. Experiments are carried out to test the performances of classification and clustering of the proposed algorithm. The results show that DualPOS is valid for attribute selection in semi-supervised learning.