Fig. 1

Flow chart of the trial. All data were split into the development and validation groups at a ratio of 7:3. The machine learning methods were performed with the training group data using a nested resampling method, in which the inner loop data were split at a 7:3 ratio with 1000 iterations to result in the best hyperparameter. Then, the best hyperparameters were used to train the machine model and tested in outer loop data with a 5-fold cross validation (CV), which resulted in a balanced performance evaluation. The machine learning that had the best performance evaluation was selected to establish the prediction model with the development data, before evaluating the prediction model with the validation data