Blending

Information Leakage

Risk in stacking where the meta-model is trained on the same data as the base models, leading to overfitting which blending seeks to avoid via hold-out validation.

← Terug