The occurrence and development of tumor is a complex and multi-stage process. Usually it's because of some gene mutation and abnormal expression or further affecting other genes'expression, which results in the change of protein molecules within cells and produces the tumor differences in the pathology and different classification in the clinical diagnosis. Therefore, the challenge of treatment in tumors is providing different patients with appropriate treatment methods, in order to get the best efficiency and do the least harm to patients. However, classification is a key question to tumor gene expression data sets.The tumor gene expressing datasets are researched in this paper. Two new methods of feature genes selection are proposed according to the data characters and biology mechanism. The main work is as follows: firstly, Kmeans_iic method is proposed, and applied to the datasets of colon cancer, prostate cancer and MLL to select feature genes .This method gets a better result and important feature genes from a great number datasets of gene chips, and it has important reference to the clinic diagnosis and biomedical research of disease. Secondly, Relief_AGA_SVM is proposed. It combines adaptive genetic algorithm with pattern recognition together to extract feature genes and retrieves the convergence fault of GA. The results showed that it can get a better effect for selecting feature genes. |