Cross-Validation

Cross-validation evaluates a model across several train-and-validation splits rather than relying on one partition. In k-fold cross-validation, the dataset is divided into k groups; each group serves as validation data once while the others are used for training, and the scores are combined. This produces a more stable estimate when data is limited and supports model or hyperparameter comparison. The split strategy must match the application: random folds can leak information when records share users, locations, subjects, or time periods. Preprocessing and feature selection must be fitted inside each training fold, and the untouched test set should remain separate for the final assessment.