Train-Validation-Test Split
← Back to Model Evaluation
Dividing data into separate sets for training, tuning, and final evaluation. Prevents overfitting to evaluation data and gives honest performance estimates.
Key Properties
- Holdout method (e.g., 70/15/15 split)
- Cross-validation: k-fold, stratified k-fold
- Time-series splits for temporal data
Related
- Bias-Variance Tradeoff (validation reveals overfitting)
- Hyperparameter Tuning (uses validation set)