Fig. 1From: Sparsifying machine learning models identify stable subsets of predictive features for behavioral detection of autismHeatmap of features used on the different folds for module 3. The darker the color of the cell, the more the feature was used in the different folds of the feature selection cross-validation (CV). Classifiers were sorted along the y-axis such that those with the highest AUCROC function were at the top. Color intensity of each cell denotes how often that feature was selected in all folds of the feature selection CV for that model. The top figure (a) used L 0 regularization, and the bottom one (b) did not. Both used the one standard error ruleBack to article page