KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
K-Fold Cross-Validation
Technique dividing data into K equal partitions to train and test the model K times in a rotating manner.
Leave-One-Out Cross-Validation
Extreme variant of K-fold where each observation serves once as test and N-1 times as training.
Stratified Cross-Validation
Preserves class proportions in each partition, essential for imbalanced data.
ROC Curve and AUC
Graphical metric evaluating the performance of binary classifiers based on true/false positive rates.
Confusion Matrix
Summary table of correct and incorrect predictions to evaluate classification performance.
Bootstrap Validation
Resampling technique with replacement to estimate model variability and performance.
Regression Metrics
Set of indicators (MAE, MSE, RMSE, R²) measuring the accuracy of regression model predictions.
Learning Curve
Diagnostic tool analyzing performance evolution based on training data size.
Temporal Cross-Validation
Adaptation respecting the chronological order of data to evaluate time series models.
F1-Score Metrics
Harmonic mean between precision and recall, ideal for imbalanced classification problems.
Negligent Cross-Validation
Double cross-validation avoiding overfitting during hyperparameter and model selection.
Validation Curve
Graph exploring the impact of hyperparameters on model performance for optimal tuning.
Group Cross-Validation
Technique preventing information leakage by grouping related observations into the same partitions.
Precision-Recall Metrics
Complementary indicators evaluating the relevance of positive predictions and their completeness.
Block Cross-Validation
Specialized approach for structured data (temporal, spatial) preserving local dependency.