Figure 9: Unbiased bootstrap tests performed on various sizes of low and high Urinary Ni samples ranging from 16 to 20, and various signature sizes ranging from 1 to 6. The statistic used for bootstrapping is combined AUC. The combined AUC is based on projection of features by multivariate logistic regression. The resampling was based on 500 permutations, as opposed to resampling with replacement, the former being more conservative. The criterion for selection of the best normalization method and the best sample cutoff value was the Achieved Significance Level of the test (ASL). The experiments were performed for several normalization methods but the diagram shows only the quantile normalization, which has delivered the best performance. As shown, the best result was obtained for cutoff value of 18, which yielded ASL < 0.01, thus giving very strong evidence against the null hypothesis. This cutoff value is used throughout this report.