An experimental comparison of Random Projection ensembles with linear kernel SVMs and Bagging and BagBoosting methods for the classification of gene expression data

In this work we experimentally analyze ensemble algorithms based on Random Subspace and Random Plus-Minus-One Projection, comparing them to the results obtained in literature by the application of Bagging and BagBoosting on the same data sets used in our experiments: Colon and Leukemia. In this work we concentrate on the application of random projection (Badoiu et al., 2006) ensemble of SVMs, with the aim to improve the accuracy of classification, both through SVMs that represent the state-of-the-art in gene expression data analysis (Vapnik, 1998) (Pomeroy et al., 2002), and through the ensemble methods, used in our work to enhance the classification accuracy and capability. Ensemble methods, in fact, train multiple classifiers and combine them to reduce the generalization error of the multi-classifiers system. To make possible the comparison of our results with those obtained in literature by the application of Bagging and BagBoosting, in this works we concentrate on SVMs with linear kernel.