Fig. 4

ROC curve comparison between Akaike Information Criterion (AIC) selected and all Maternal Frailty (MaFra) risk factors. We compare the ROC curves of a random forest [95] trained only on the subset of 9 risk factors determined by AIC in the explanatory phase against a random forest trained on all the MaFra risk factors. We estimate the classification error for each competitor by means of leave-one-out-cross-validation (i.e. LOOCV see ch. 7 in [96] for a detailed explanation). We present also the area under the curve specific to each competitor in different colors (i.e. orange for the 9 risk factors selected by AIC and blue for the whole set of MaFra risk factors)