KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Level 2 features
Transformed variables representing the predictions of base models used as inputs for training the meta-model.
Blend stacking
Simplified variant of stacking using a single hold-out validation set to generate meta-model predictions.
Multi-level stacking
Extension of stacking that stacks multiple layers of meta-models to capture complex interactions between predictions.
Stacking overfitting
Increased risk of overfitting due to the increase in model complexity by adding meta-learning.
Adaptive weighted aggregation
Meta-model mechanism that dynamically learns optimal weights for each base model according to regions of space.
Cascade modeling
Stacking architecture where successively trained models correct the errors of previous models.
Probabilistic stacking
Bayesian approach to stacking where the meta-model combines probability distributions rather than point predictions.
Meta-prediction calibration
Process of adjusting the meta-model's probability scores to ensure their statistical reliability.
Hybrid stacking
Combination of stacking with other ensemble techniques like bagging or boosting to maximize diversity.