AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Level-0 Learner
Refers to the base models in stacking, which perform initial predictions on the data before their combination by the meta-model.
Level-1 Learner
Synonym for meta-model, it is the second learner in stacking that takes as input the predictions of level-0 learners to produce the final prediction.
Feature Stacking
Technique where base model predictions are concatenated with original features to form the input set for the meta-model, thus enriching the information.
Data Leakage
Critical problem in stacking if predictions on the training set are generated by models already trained on these same data, leading to overestimation of performance.
Stacked Cross-Validation
Rigorous procedure for training a stacking architecture, integrating cross-validation both for training base models and generating predictions for the meta-model.
Heterogeneous Models
Fundamental principle of stacking that involves combining models of different natures (e.g., decision tree, SVM, regression) to capture various patterns in the data.
Cascade Training
Training strategy for stacking where base models are trained sequentially, each new model potentially using predictions from previous models as additional features.