Research On Information Gene Selection Algorithm Based On Mutual Information

Mutual information is an important concept in information theory,which is used to measure the correlation between two random variables.The Maxi-mum Information Coefficient(MIC)and the Maximum Relevance&Minimum Redundancy(mRMR)based on the mutual information theory are the two most representative methods in feature selection.In view of the characteristics of small sample,high dimension and high noise of tumor gene expression profile data,it is very important to select a suitable algorithm to select the information gene.Although MIC has the advantages of generality and equitability,it can only mea-sure the correlation between feature and category in the selection of information gene,and can not effectively remove the redundancy between features.Although the mRMR can effectively remove redundant features,it has some limitation-s to the data size.A new algorithm,mRMR-ChiMIC algorithm,is proposed to improve mRMR in this thesis.We replace the original mutual information algorithm by the normalized mutual information(MIC)in the original mRM-R algorithm,and the ChiMIC algorithm is used to approximately estimate the value of MIC.We test the new algorithm on three common data sets of DLB-CL?Prostate?Lung,the experimental results show that the mRMR-ChiMIC algorithm has higher classification accuracy and lower computational complexity compared with the original mRMR algorithm.