Research On The Gene Selection Based On Rough Sets Theory

Rough set is a new mathematic tool for dealing with fuzzy and uncertain knowledge, as a new method for acquiring knowledge, inaccurate, inconsistent, incomplete information can be effectively analyzed. Different from other methods dealing with uncertain and imprecise question, prior information is not necessary, which describe the problem in objective.In view of the gene expression data has many characteristics such as high dimentional,nonlinear and high noise.Rough set theory can effectively deal with the gene expression data based on the advantages of tiself.Rough set theory is applied to selecting gene is a cutting-edge research and a development research subject.Based on the research of rough set theory, we apply it to gene selction, two gene selection methods is got.(1) We can obtain the effects of each attribute inputting to the decision-making and the dependable relationship between the attributes through the method of analysis of attribute dependability in rough sets. In order to get the atrribute which influenced the decision-making mostly,a novle method called Maximum Dependency of Attributes Based on Rough Sets(MDA-RS) was proposed.Here the MDA-RS method was applied to gene selection.First heuristic k-means clustering algorithm was applied to analyze these genes for clustering.Then a representative gene was selected from the clusering gene using MDA-RS. Then the set of representative genes was considered classification features.(2) Gene selection based on rough-genetic algorithm(RGA),which is conduced by combined rough set theory with genetic algorithm. Here the RGA method was applied to gene selection.The main content is that initiating the population with encoding individual with kernel character as limits;Conducting variation factor with the significance of rough set;Adustment operator is added to the modified method, which improve the fitness of individual to some extent.Experimental results with the public dataset suggest that the propsed methods not only can confirm the most informative gene subset but also improve the classification accuracy.The methods are feasible and effective.