Abstract:

Rough set theory is an effective method  of  knowledge reduction, but can not directly deal with numerical attributes. To address this problem, based on the concepts of neighborhood and cover, the algorithms of covering itself reduction and covering reduction are constructed. And then, by discussing the relationship among neighborhood samples, the mutex in neighborhood samples is defined. The unreasonable positive region caused by mutex, leads to  attribute dependency that may not accord with what the real condition dataset reflects, thus the method of decompose mutex neighborhood is proposed. Finally, we report experimental results with four numerical gene expression datasets and compare the results with some other methods. The results prove that the proposed method is effective, and has higher tumor classification accuracy.

Key words: mutex; neighborhood; covering rough set; attribute reduction