Cross-Validation
Train and test on multiple splits of the data to get a more reliable performance estimate.
- K-fold rotates through K splits and averages the score.
- K=5 or K=10 are standard.
- A single split can be lucky; CV tells you how stable the model is.
When to use: For hyperparameter tuning, model selection, and whenever you need a stable performance estimate.
Example: 5-fold CV on a 10k-row dataset: train on 8k, test on 2k, rotate 5 times. Report mean ± std test score: 0.84 ± 0.02 AUC.