The Effects of Variable Selection and Dimension Reduction Methods on the Classification Model in the Small Round Blue Cell Tumor Dataset

Abstract

Objective: The purpose of this study is to investigate and compare the effects of different dimension reduction methods (PCA, ICA, PCA + Forward Selection, ICA + Forward Selection) on the K-NN classifier using open access gene expression data of small round blue cell tumor types.Methods: In this study, open access gene expression data of small round blue cell tumor types was used for investigate and compare the effects of different dimension reduction methods. In the study, PCA, ICA, PCA + Forward Selection, ICA + Forward Selection were used as different dimension reduction methods together with K-NN classification method.Results: Accuracy values obtained from the dimension reduction model made with PCA on K-NN model; for EWS, BL, NB, and RMS type tumors with 93.51%, 91.14%, 92.31%, and 94.74% respectively. Accuracy values obtained from the dimension reduction model made with PCA + Forward Selection on K-NN model; for EWS, BL, NB, and RMS type tumors with 96.25%, 96.25%, 95.06% and 95.47%, respectively. Accuracy values obtained from the dimension reduction model made with ICA on K-NN model; for EWS, BL, NB, and RMS type tumors with 91.89%, 90.67%, 88.31% and 89.47% respectively. Accuracy values obtained from the dimension reduction model made with ICA+ Forward Selection on K-NN model; for EWS, BL, NB, and RMS type tumors with 93.51%, 91.14%, 92.31% and 94.74% respectively.Conclusion: In this study, the model created with PCA gives higher results than the model created with ICA. In addition, according to the results of the models obtained by applying the Forward selection method on these 2 models, the forward selection method has increased the classification performance.