Figure 2

Misclassification error as a function of the regularization parameter computed with nested cross-validation. (a) Murine data experiment: The misclassification error reaches a plateau after α = 110, and the \({\ell }_{2}\) parameter γ has no influence. (b) Alzheimer human experiment: The misclassification has two plateaus, one near α = 20 and one near α = 190; for the \({\ell }_{2}\) parameter, γ > 0.03. (c) ADHD human experiment: The misclassification has a plateau near α = 10; the \({\ell }_{2}\) parameter changes the results but with minimal influence. (d) The number of features detected by one algorithm and not by the other varying the amount of sparseness. This graph shows that by decreasing the sparseness, the number of features detected by the MLA increases. The example shown is for the Alzheimer dataset.