Fig. 2From: Sparsifying machine learning models identify stable subsets of predictive features for behavioral detection of autismHistogram of predicted probabilities of the two discarded models on the cross-validation set. The x-axis denotes the predicted probabilities and the y-axis the number of subjects. The top figure (a) corresponds to AdaBoost and the bottom figure (b) to Gradient Boosting. AdaBoost probability plot is unstable, and gradient boosting has a high rate of misclassification among the control subjectsBack to article page