التحقق المتقاطع K-Fold

A single train/test split gives just one score. If that split was lucky or unlucky, your estimate could be misleading.

Cross-validation reuses your data many times, testing on a different slice each round, then averages the scores for a steadier estimate.

You chop the data into k equal parts called folds. A common choice is five folds, giving five separate test rounds.

Each round, one fold becomes the test set and the other k minus one folds train the model. Then the test fold rotates to the next one.

Over k rounds, every row is tested on exactly once and trained on the rest of the time. No data is wasted. 🔄

You end with k scores, one per fold. Their average is your headline estimate, and their spread shows how stable the model is.

scikit-learn does the looping for you. The helper cross_val_score returns one score per fold in a single line.

The returned array holds each fold's score. Take its mean for the headline and its standard deviation to gauge reliability.

Five or ten folds are typical. More folds train on more data per round but cost more compute, so it is a speed-versus-stability trade.

For classification, use StratifiedKFold so each fold mirrors the overall class balance. scikit-learn applies it automatically for classifiers.

Cross-validation guides model choice during development. Still keep one untouched test set aside for a single honest score at the very end.

In 5-fold cross-validation, how often is each row tested?

Split into k folds, rotate the test fold, and average the scores for a robust estimate. Use cross_val_score, then a final hold-out. 🔄