Fig. 5. The proposed framework for developing an intelligible TBI prognostic model.

After training an initial machine learning model, input features are selected based on statistical robustness and clinical validity. The machine learning model is retrained after each step of feature selection.