Figure 2.

Importance the five feature categories for the DISOselect's models designed for the 12 disorder predictors. We used a three‐step process to derive the scores for each model. First, the information gain of individual features was calculated from the extra‐tree regressors. Second, features were divided into the five classes and the information gain of the features in the same category was summed up. Third, the summed values were dived by the sum of the information gain values of all features in the same model. The last step allow for directly comparison of relative contributions of each feature category