Feature Generation For Level 1 Models

Level 1 features are the predicted probabilities for Class_1 claim. Generating these features required partitioning the Kaggle training data set into mutually exclusive sets to avoid over-fitting brought on by using the same instances of training data for both Level 0 and Level 1 models.

For this competition a 5-fold partition of the Kaggle training data was used to create Level 1 features. This approach is illustrated by Faron's reply in this forum posting.