Fuzzy and Rough Set Theory Based Gene Selection Method

The selection of genes from microarray gene expression datasets has become an important research in cancer classification because such data typically consist of a large number of genes and a small number of samples. In this work, Neighborhood mutual information is retrieved to evaluate the relevance between genes and is used to stop information loss. Firstly, an improved Relief Feature Selection algorithm is proposed to create candidate subsets of features. Based on the neighborhood mutual information the cohesion degree of neighborhood object and coupling degree between neighborhood objects have been defined. Furthermore, a new method of initialization for cluster centers in Fuzzy C-means (FCM) algorithm is proposed. FCM allows one piece of data that belong to two or more clusters. Neighborhood rough set is used for extraction and selection of features and is used in proposed FCM algorithm. Finally, to find the performance of the proposed approach, five gene expression datasets were taken. Experimental results show that the proposed approach can select genes effectively, and can obtain high and stable classification performance.