Classification And Gene Selection Of Breast Cancer Based On Genetic Algorithm And Weighted Extreme Learning Machine

There are various method to solve this feature select problem.What we choose is a kind of Wrapper.That is search for a best set of genes in the space of possible genes set.However,here comes tow problem.One is how could we define “good” set of genes and the other is how could we find the best set of genes.For the first problem,we solved it as we model the data represented by the gene set then predict subtypes by these models.The more precisely that the model predict,the better the set of genes is.We choose extreme learning machine as the model because it is easy,extreme fast and very suitable for large-scale and repeate train.For the second problem,there exist lots of search methods such as random search,exhaustion,ant colony algorithm and so on.We choose genetic algorithm because it is relatively easy and robust.This paper also draws on other feature selection method which is filter method.By SAM method and similarity-based greedy algorithm we filter out a large number of redundant genes,reduces the search space,and removed many useless genes for the interference of the algorithm.We call this process genetic screening.After that,for the problem and data,this paper gives the genetic operations and involving various parameter Settings that are adapted to solve the problems,.This completed for this kind of problem of a kind of brand-new worthy of reference solution.In the study,we have a genetic data imbalance problems,through the analysis and research,we use the weighted extreme learning machine and solved such problems by means of price sensitive.Finally,We gives the results of our method compared with other methods.We first compared the performance of the different machine learning algorithms on the genes we get,which illustrate the selected gene set is robust.Different learning algorithms all get more than 95% geometric average performance prediction accuracy.Then we compared the prediction performance of the gene sets selected by our method with those obtained by other methods.Our method takes 96.53% of the geometric average prediction accuracy,which is better than any other method and proves that our method is worthy of study and expansion.