Research Of Tumor Gene Analytic Method Based On 2DEPCA And RGS-SVM

Tumor is one of the main causes of today's threats to human life.To all researchers,the key focus is how to identify and diagnose tumor.With the development of biological information technology,extracted the disease genes form gene expression profile has become the main mean of tumor treatment and prevention.Feature extraction is a very effective method in data mining,but the spectrum of gene expression in gene expression data often have high dimension,high noise and high redundancy characteristics,greatly reduces the performance of feature extraction method,and lead to the complexity of the sharp rise in method.Therefore,to design an effective feature extraction method is particularly important in tumor diagnosis;In addition,in the process of gene classification,the choice and construct classifier are also great influence on the classification results,and a good classification model can not only improve the classification accuracy,but also can improve the efficiency of classification.Therefore,in this paper based on the process of genetic classification,the feature extraction and classification model are improved respectively.The main research work is as follows:(1)The traditional feature extraction method is generally not make full use of the correlation between genetic information and structural features,which leads to the low classification of the overall performance.Therefore,this paper has a try to introduce the 2DPCA algorithm into feature gene selection which method used in the field of the image recognition,and aiming at the high redundancy defect of 2DPCA,a based on the information entropy of 2DPCA feature extraction method(2DEPCA)is proposed.The basic idea of this method can be described as: firstly,the entropy of information of thermodynamic statistical is introduced into feature subset primaries of gene expression data,and then the characteristics genes after the primaries are extracted by 2 dimensional principal component analysis.The experimental result demonstrates that the 2DEPCA can improve classification accuracy and reduce redundancy when compared to other related methods.(2)For the low efficiency of finding parameters of SVM training,based on the traditional gird search method a based on the SVM adaptive tumor classification method(RGS-SVM)is proposed.The basic idea of this method can be described as: by the thought of binary and search and iteration,the search range is reduced and the search efficiency is improved.The six groups are used to test our method,and the experimental result shows that the RGS-SVM method can not only improve classification accuracy and reduce redundancy when compared to other related methods.