Dimensional Reduction Research And Application On The Cancer Gene Data

Cancer has always been one of the most deadly diseases by now. Thanks to the rapiddevelopment of gene chip technology, researchers have gained vast amounts of cancergene expression data. Using gene expression data to diagnose the disease has become a hottopic of the post genome era. The accuracy of Gene expression data classification will helpto improve the efficiency of diagnosis. However, gene expression data generally aresample sized, high dimensional, non-linear.According to the basic characteristics of gene expression data, this paper apply theSVM (Support Vector Machine) to classify the gene data. SVM is a kind of machinelearning theory based on statistical learning theory, which can be applied to this situationwhere the number of gene feature is much more greater than the purpose of samples. Beingdifferent from the ERM (Empirical Risk Minimization) principle, SRM (Structural RiskMinimization) rule is to seek good generalization ability decision function, effectivelyavoids the local optimal solution. Under the limited training samples, getting the decisionrule for an independent test set via SVM still can reach a small error.Although SVM’s training strategy is good enough to avoid the local optimal, thesmall-sample trouble still exists. In order to get more accurate results, dimensionalreduction algorithm will be considered in this paper. The dimensional reduction algorithmswill be PCA (Principle Component Analysis), MDS (Multi-Dimensional Scaling),Laplacian Eigenmaps, etc.In this paper, two common data sets, Lung and DLBCL, will be used. Theexperimental results show that the dimensional reduction algorithm based on SVMattribute a significant effect to the optimization of the classification accuracy.