Very interesting! Do you have any literature on how to prevent overfitting the meta-model? Or how to detect if I overfit the primary model or the meta model?
1) It's quite easy to detect overfitting in both cases: it occurs when you have excellent performance on the training set but poor results on unseen data.
2) Preventing overfitting in the meta-model works like with any other ML model (avoiding bias, using cross-validation, etc.). You just need to be especially careful about data leakage from the base model and the typically small number of observations.
Here is the link of two very interesting books from Marco Lopez de Prado, they took more in details about meta-labeling (just look at the table of contents).
Very interesting! Do you have any literature on how to prevent overfitting the meta-model? Or how to detect if I overfit the primary model or the meta model?
Hi Jacob,
1) It's quite easy to detect overfitting in both cases: it occurs when you have excellent performance on the training set but poor results on unseen data.
2) Preventing overfitting in the meta-model works like with any other ML model (avoiding bias, using cross-validation, etc.). You just need to be especially careful about data leakage from the base model and the typically small number of observations.
Here is the link of two very interesting books from Marco Lopez de Prado, they took more in details about meta-labeling (just look at the table of contents).
https://agorism.dev/book/finance/ml/Marcos%20Lopez%20de%20Prado%20-%20Advances%20in%20Financial%20Machine%20Learning-Wiley%20%282018%29.pdf
https://agorism.dev/book/finance/ml/Machine%20Learning%20for%20Asset%20Managers%20by%20Marcos%20M.%20López%20de%20Prado%20%28z-lib.org%29.pdf
Hope it helped. Have an excellent day!