Figure 1. Design of Machine Learning Algorithms.

The level 1 classifiers consist of 3 independent models each trained on the same initial training sample (sample A), including logistic regression with least absolute shrinkage and selection operator (LASSO), extreme gradient descent boosting (XGBoost), and a neural network. The next training sample (sample B) is then input into the level 1 classifiers, resulting in 3 risk estimates for each observation in sample B, 1 from each level 1 model. These 3 risk estimates are then used to train the level 2 XGBoost classifier (sample C). A final sample (sample D) is input into the level 1 classifiers to obtain risk estimates for input into the level 2 classifier. Performance of the level 1 and level 2 classifiers is assessed using this final training set D.