What is min_samples_leaf in Decision Trees?
Updated May 16, 2026
Short answer
It sets the minimum number of samples required in a leaf node.
Deep explanation
min_samples_leaf ensures that each terminal node contains enough samples to make reliable predictions. This smooths the model by preventing leaves that represent outliers or noise. It is particularly useful in noisy datasets where overfitting is a risk.
Real-world example
Used in fraud detection systems to avoid flagging rare random behavior as fraud.
Common mistakes
- Confusing it with min_samples_split.
Follow-up questions
- Does increasing it reduce overfitting?
- Can it affect bias?
- What is a good default value?