Unimodal transform of variables selected by interval segmentation purity for classification tree modeling of high-dimensional microarray data

Leukemia and small round blue-cell tumor cancer detection using microarray gene expression data set: Combining data dimension reduction and variable selection technique

“…This subject indicates that the ability of BSS/WSS for identifying signification genes is restricted. Finally, the proposed GA-LDA based dimension reduction method produces prediction accuracy similar (somehow better than in calibration) to that of the novel variable selection-multivariate calibration method (unimodal transform of variables selected by interval segmentation purity; UTISP-based CART) [34] method; however the latter produces a more accurate prediction in test set. Nevertheless, the mathematical description of UTISP-based CART is complex with respect to GA-LDA based dimension reduction.…”

“…The comparison between the GA-LDA based on dimension reduction model for classifying different kinds of small, round blue-cell tumors (SRBCTs) and the previously presented model is reported in Table 6. The prediction accuracy provided by the proposed method is higher than those of k-NN, CART, BSS/WSS based CART and even UTISP based CART [34]. Among these methods, k-NN and CART uses the information of all genes variable and the redundant parts of the data lead to lower the prediction accuracy of these model.…”

“…As it can be seen in Table 6, GA-LDA based on dimension reduction can provide models with prediction abilities better than k-nearest neighboring (k-NN) and conventional classification and regression tree (CART) model. As discussed by Du et al [34], using irrelevant or redundant genes variables by k-NN and CART, increase the risk of over-fitting of the classification model. Thus, variable selection based on ranking the ratio of between classes sum of squares to within class sum of squares (BSS/WSS) has been employed to select the relevant variables in the Leukemia data set.…”

“…Among these methods, k-NN and CART uses the information of all genes variable and the redundant parts of the data lead to lower the prediction accuracy of these model. However Qin Yu et al [34], increased the prediction power of CART by its combination with novel variable selection, which has been named unimodal transform of variables selected by interval segmentation purity; UTISP.…”