Fig. 1

The leave-one-database-out design used to evaluate the transportability of the Level 1 models trained using a single database and the Level 2 ensembles that combine multiple Level 1 models. The figure shows that five different combinations were used, where four of the five databases were used to develop the models and the final database was used to fairly evaluate the transportability of the models. In addition, a model was trained using the left-out database to calculate the internal validation that could be considered the ‘internal benchmark’ performance for the database given sufficient training data. We compared how similar the external validation of each model was with the ‘internal benchmark’