KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Stratified K-fold
Version of K-fold that preserves the class distribution in each partition, essential for unbalanced datasets in classification.
Time Series Cross-Validation
Technique adapted to time series data using successive time ranges as test sets without mixing past and future observations.
Nested Cross-Validation
Double cross-validation where an inner loop optimizes hyperparameters and an outer loop evaluates the performance of the optimized model.
Group K-fold
Variant of K-fold ensuring that the same groups never appear simultaneously in training and test sets.
Shuffle Split Cross-Validation
Method that randomly generates training/test partitions with a configurable number of iterations and set sizes.
Repeated K-fold
K-fold repeated several times with different random initializations to reduce the variance of performance estimation.
Holdout Validation
Simple method separating data into a single training set and a single test set, less robust than cross-validation.
Cross-Validation Score
Average performance metric calculated on all cross-validation partitions, often with its standard deviation to measure stability.
Grid Search Cross-Validation
Exhaustive search of hyperparameters combined with cross-validation to identify the best model configuration.
Randomized Search Cross-Validation
Alternative to Grid Search that randomly samples hyperparameter combinations with cross-validation to optimize computation time.
Cross-Validation Folds
Individual partitions of data created during cross-validation, serving alternately as test or training sets.
Monte Carlo Cross-Validation
Cross-validation method that randomly repeats the training/test split multiple times to estimate the performance distribution.
Cross-Validation Iterator
Object that generates partition indices for cross-validation, implementing different data splitting strategies.
Adaptive Cross-Validation
Advanced technique that dynamically adjusts the validation strategy based on data and model characteristics.
Cross-Validation Leakage
Information leakage between training and test sets due to incorrect preprocessing, invalidating cross-validation results.
Bootstrap Cross-Validation
Method that uses sampling with replacement to create validation partitions, offering a different estimate of generalization error.
Prequential Cross-Validation
Validation strategy for data streams that tests each observation immediately after its learning, adapted to evolving concepts.