Miks töötab bagging puude puhul, nagu juhuslikus metsas?

Bagging (Bootstrap Aggregating) works well with decision trees because:

1. High model variability: Individual decision trees tend to overfit and have high variance. Bagging reduces this variance by averaging the predictions of many trees trained on different data subsets.

2. Model independence: Since each tree is trained on a random sample with replacement (bootstrap), the models are sufficiently different, improving the averaging effect.

3. No need for strong base models: Bagging works better with unstable models (like trees), where small changes in data significantly affect the model structure.

4. Simplicity and efficiency: Trees are easy to train and combine, making bagging practical and effective.

In random forests, additionally, a random feature selection is applied during node splitting, which further increases the diversity of trees and improves the model's generalization ability.