Feature extraction from tumor gene expression profiles using DCT and DFT

Feature extraction plays a key role in tumor classification based on gene expression profiles, which can improve the performance of classifier. We design two novel feature extraction methods to extract tumor-related features. One is combining gene ranking and discrete cosine transform (DCT) with principal component analysis (PCA), and another is combining gene ranking and discrete Fourier transform (DFT) with PCA. The proposed feature extraction methods are proved successfully and effectively to classify tumor dataset. Experiments show that the obtained classification performance are very steady, which are evaluated by support vector machines (SVM) and K-nearest neighbor (K-NN) classifier on two well-known tumor datasets. Experiment results also show that the 4-fold cross-validated accuracy rate of 100% is obtained for the leukemia dataset and 96.77% for the colon tumor dataset. Compared with other related works, the proposed method not only has higher classification accuracy rate but also is steadier in classification performance.