Fig. 5: Dot plot of feature importance.

Dot plot of feature importance calculated using the mean SHAP values from running 10-fold cross validation on the training set for 1000 times. The x-axis was the SHAP value (in unit of log odds) of the features used for diagnosis classification. Each row shows the importance of different values of one feature, with red color indicating high feature values and blue color indicating low feature values. The positive x-axis represents the importance of each feature value to support unipolar depression, and the negative axis represents the importance of each feature value to support bipolar depression. The rows were ranked by the overall feature importance vertically. Therefore, disease duration has the strongest drive to the model’s prediction, while LYMPH ratio has a relatively weaker drive. Notably, when points do not fit together on the line, they pile up vertically to show density.