Research Of DNA Microarray Data Classification Based On SVM

In the late 20 th century,the rapid development of information technology gave birth to a new discipline called bioinformatics.Depending on the organic combination of mathematics and AI(artificial intelligence)with biomedicine,Bioinformatics has made fruitful achievements in tumor gene expression profile analysis,gene variation,protein structure analysis,etc.,and has become a hot research issue.The use of information processing technology for the analysis of tumor gene expression profile data is of great practical significance in revealing the causes,development mechanism and diagnosis of diseases and in drug development.Gene expression profile data has its own characteristics,which were mainly expressed in small samples,high-dimension,high noise and high redundancy.These characteristics result in a poor effect when gene expression profile data is processed by using traditional classification methods;however,a good effect can be obtained by using a support vector machine(SVM)for the classification of small-sample and high-dimension data.Therefore,it is necessary to select features from original data in order to overcome the effect of high noise and redundancy of gene expression profile data on the performance of classifiers.In combination of the existing methods commonly used in gene expression profile data classification,this paper carries out research mainly in the following two aspects regarding the problems in these methods:(1)feature gene selection method.First,the ReliefF method is used to screen feature genes so as to reduce the dimension of search space and to remove redundancy and noise;second,a hybrid harmony difference algorithm is used to determine an optimal feature gene combination.The hybrid algorithm can overcome not only the shortcomings of the traditional harmony search method,namely poor local optimization capacity and low solution precision,but also the problem that differential evolution algorithm is easy to fall into local optimum.The result of a simulation test shows that the use of the hybrid algorithm for feature gene selection has good performance in optimization of accuracy,stability,etc.(2)Research of gene expression profile classification based on improved SVM.Determining all parameters using the SVM often depends on experience.This paper uses the improved DE algorithm to find optimal parameters by converting the problem of determining the parameters(penalty parameters and width coefficients)of radial basis function SVM to a combination optimization problem.Dynamic parameter adjustment strategy is used to improve the performance of the differential evolution algorithm;and an elite substitution strategy is added to enhance the convergence of the algorithm.The result of a simulation test shows that the improved DE algorithm can effectively improve the accuracy of SVM classifier,and has a good generalization performance.