Fig. 5

The permutation feature importance of the machine learning models. a-c The 10 most important features were ranked, and their boxplots show the distribution of the decrease in the AUROC score. The lower the saturation, the higher the importance score of the feature. d The 10 most important features were analyzed with the coefficient of the LR. Features related to a positive DILI prediction are shown in blue, and those related to a negative DILI prediction are shown in red. The lower the saturation, the higher the importance score of the feature. Large absolute means that feature is important