Fig. 2.

Machine learning analysis to predict dementia based on centrality metrics using the ADNI data set. (A) The mean performance profile computed as a function of the number of features included in the training process of radial SVMs (with the parameter $C=100$ controlling the complexity of the decision boundary). The most relevant features are identified at smaller values of the x-axis, so features were added progressively according to their relative importance. The max AUC, black diamond, was achieved with the first 86 most important features. Supplementary Fig. 1 shows the histogram of the AUC index at this point. (B) The mean ROC curve generated from the features identified in (A). Sensitivity, specificity, and accuracy indices were computed at the optimal point (red circle). Such a point was identified as the closest point (in terms of Euclidean distance) on the ROC curve to the point defined by a true positive rate of 1 and a false positive rate of 0. Abbreviations: ADNI, Alzheimer's Disease Neuroimaging Initiative; AUC, area under the curve; ROC, receiver operating characteristic; SVM, support vector machine.