The effect of noisy bootstrapping on the robustness of supervised classification of gene expression data

This paper discusses the role of noisy bootstrapping in the analysis of microarray data. We apply linear discriminant analysis, according to Fisher´s method, to perform feature selection and classification, creating a linear model which enables clinicians easier interpretation of the results. We present the effects of bootstrapping in the improvement of the results, and specifically robustifying classification with an increased number of genes. The performance of our method is demonstrated on the publicly available datasets, and a comparison with state of the art published results is included. In particular, we show the effect of the number of features (genes) on the result, as well as the effect of bootstrapping. The results show that our classifier is accurate and quite competitive to other classifiers, although it is simpler, and enables considering a larger set of genes in the classification