Fig. 1.

Flowchart of the study design. In step 1 of model selection, we performed training-testing splitting 100 times on the discovery set and trained DT, RF, GB models on the training sets, then compared these results with FIB-4 on testing sets. In step 2, we constructed final GB models and compared results with FIB-4 on the whole discovery set and then validated on the HBV validation set. In step 3, GB models and FIB-4 were used to predict the risks for two extra HCV cohorts. In step 4, we developed a user-friendly web-tool for clinical practices.