A Study Of Gene Selection Method Based On Scoring Criterion And Particle Swarm Optimization

Cancer is a life killer due to its complexity and wide varieties.It is critical to adopt appropriate treatment according to specified diagnosis for each cancer variant needs specified treatment.The emergence of gene chip provides a new path for human beings to understand the mechanism of disease at the angle of molecular.It is great significance to find out the disease-causing gene for the diagnosis and treatment of cancer by data mining of gene expression profile.Many ways has been proposed for gene selection which can select out the gene subset with a high classification accuracy,but most of them are complex,time-consuming,poor interpretative and redundant.In order to overcome the shortcomings of these methods,on the basis of scoring criterion,particle swarm optimization algorithm and the extreme learning machine are used in this paper to select genes.The main work of this paper is shown as follow:(1)A new method of gene selection based on scoring criterion and improved PSO algorithm is proposed to reduce overhead in traditional methods.Firstly,the original gene pool is preprocessed by using the information index to classification(IIC),then the gene set matrix is established randomly according to the scientific random sampling,the gene set is evaluated by using the extreme learning machine,moreover,the gene set is select out which satisfies the condition.Then,the proposed scoring criterion is used to evaluate and sequence each gene so that the irrelevant genes are filtered;Finally,the PSO algorithm is improved by using simulated annealing to select the genes any further.The new method has simple steps and low cost.The experiment result on a number of open data sets shows that compared with other method,the subset of genes with high classification accuracy can be selected quickly and efficiently due to accurate remove of a large number of redundancy by using the proposed method.(2)Since the scoring criterion cannot make full use of the genetic information and the PSO algorithm is still easy to fall into the local optimal,information weighting and particle half-initialization are proposed to improve the method.Firstly,according to the variance,the average times of fitness value calculation is adjusted,and then the classification weight information contained in the gene itself is used as the new standard of the scoring criterion to improve the scoring mechanism.Finally,aiming at the shortcomings of the PSO algorithm,a threshold is set to force half of the particles update to improve the algorithm.The improved method makes full use of the information contained in the gene,making the scoring mechanism more reasonable and enabling particles jump out of the local optimal easier than other method.The effectiveness of the improved method is verified by the experimental results on the four datasets.