Stratified K-Fold Cross-Validation

The Problem

Imagine your dataset has 99% Class 0 and 1% Class 1.

• Random K-Fold might accidentally put all Class 1 samples in Fold 3.
• Folds 1,2,4,5 train on only Class 0 data → They never learn the minority class!
• Results are misleading.

The Solution: Stratified K-Fold

Stratified K-Fold ensures every fold has the same class distribution as the original dataset.

How it works:

1. Sort data by class label.
2. Divide into K groups, alternating class labels.
3. Each fold contains ~1% Class 1 and ~99% Class 0.

Result: Every fold is a representative miniature of the whole dataset!