Gene selection algorithm combining ReliefF and relative neighborhood rough set

The curse of dimensionality, caused by high-dimensionality gene and small-size sample of gene expression dataset, may degrade the accuracy of tumor classification. To solve the issue, in this paper, the neighborhood rough set theory is introduced, and through expanding neighborhood threshold, the relative neighborhood rough set theory is proposed. Some corresponding theorems are drawn, and a gene selection algorithm of multi-class problem, by combining ReliefF and relative neighborhood rough set, is constructed. Finally, through comparing with other methods on four opened gene expression datasets, our method shows that only few genes could achieve higher tumor classification accuracy.