Kreuzvalidierung (EN)

Concept

Cross-validation is a statistical procedure for model validation that enables a more reliable assessment of model performance. It works by repeatedly splitting the data into different training and test sets, making the results more robust. Unlike simple division into training and test data, cross-validation uses all available data for both training and validation. This method significantly reduces the influence of random splits on model evaluation.

How it works

flowchart TD     A[Original dataset] --> B[Split into k folds]     B --> C[Iteratively for each fold i]     C --> D[Use fold i as test set]     C --> E[Use remaining k-1 folds as training set]     D --> F[Train model on training set]     F --> G[Evaluate model on test set]     G --> H[Save performance metric]     H --> I[All k iterations completed?]     I -->|No| C     I -->|Yes| J[Calculate average performance] 

In context

• Typically used together with hyperparameter optimization
• Related to: Bootstrap, Holdout method, Stratified Cross-Validation
• Example use: Evaluating the accuracy of a classification model