PERFORMANCE OF CLASSIFICATION METHODS FOR DIFFERENTIATION BETWEEN CIRRHOTIC TISSUES AND CIRRHOTIC TISSUEWITH CONCOMITANT HEPATOCELULAR CARCINOMA. CLASSIFICATION OF LIVER TISSUES

This paper presents the comparison of various discriminant methods for differentiation between cirrhotic tissue and cirrhotic tissue with concomitant hepatocelular carcinoma on the basis of oligonucleotide microarray dataset. Four methods of dimensionality reduction by selection of features (genes) subsets: linear models with empirical Bayes methods [Smyth 2004], SAM, PAM and Wilcoxon statistic were implemented. For studied subsets of genes ranked by these methods the performance of seven discriminant procedures was estimated by test error, 10-fold CV and bootstrap 0.632 with 95% confidence intervals. The best performance was obtained for SVM, bootstrap aggregating trees as well as adaptive boosting trees.