Fig. 6
From: Effective workflow from multimodal MRI data to model-based prediction
Beeswarm plots of SHAP values for the best predictors of each target. Data points are reflecting the individual subjects, where the SHAP values were averaged across 100 iterations of the nested 5-fold cross-validation for each individual. The vertical axis includes the prediction target (leftmost column), the feature type (middle column) and the brain atlas (rightmost column). Colors of the dots depict feature values of the best predictors with the scaling given on the color bar. The numbers in the plot indicate the mean absolute SHAP values (mean(|SHAP|)) reflecting the SHAP feature importance. The abbreviation ‘Emp.’ is for empirical features, ‘Sim. (Low dim.)’ and ‘Sim. (High dim.)’ are for the simulated features of the low- and high-dimensional parameter optimizations, respectively.
