Optimization Method Of SVM Classifier For Gene Microarray Data

Support vector machine(SVM)is a type of new machine learning method based on statistical learning theory, which is one of the most popular classification techniques nowadays. Compared with some other learning algorithms,SVM uses the principle of minimizing structural risk,and preferable to be used to resolve some small sample problems, especially for gene expression data with high dimension, small numbers and nonlinearity.DNA microarray chips loaded thousands of DNA fragments produce a large number of gene microarray data with a high medical applications value for diagnosis and treatment in some research and analysis. With the rapidly growing of microarray information processing and information technology development, SVM as a potential data mining technology, has become an important research direction of the study of microarray chip. The main research work about this issue is as follows:(1) For the curse of dimensionality problem of gene microarray data, this paper will design an optimization method which combines support vector machines with a variety methods of dimensionality reduction. This paper will do some research of a variety of dimensionality reduction methods, which probably affect the improvement on the performance of SVM classifier.(2)The paper shows the classification results of SVM classifier after dimensionality reduction via the use of five cancer data sets. It has a comparative analysis of linear and nonlinear dimensionality reduction methods and their affection the results of classifier, suggesting the underlying non-linear structure of high dimensional biomedical data.(3)The paper gives visualization maps of two kinds of linear methods (PCA, MDS) and three non-linear dimensionality reduction methods (Isomap, LLE, LEM) and cluster validity measures in the low-dimensional space.