Performance of three classifiers varying the between class differences and normalization methods. The figure shows the overall predictive accuracy (PA) for 1-NN, DLDA and PLR, obtained varying the difference between classes. We considered the same simulation settings of the left panels of Fig. 3 (p = 1000) but varied the mean of the p
= 20 non-null variables (μ(2)). The results obtained by normalizing samples (left panels) or variables (right panels) are reported. Details on data generation are reported in the Methods section.