Differences Between Bagging and Random Forest
📌 The primary difference is the base estimator: Bagging allows any algorithm (like Decision Trees, SVMs, etc.), while Random Forest is strictly built using Decision Trees ($\text{DT}$).
📌 When making a Bagging ensemble with only Decision Trees, it is not equivalent to a Random Forest; the distinction lies in feature sampling.
📌 Bagging performs tree-level feature sampling: the subset of features used for the entire tree is decided once at the beginning.
📌 Random Forest performs node-level feature sampling: a new random subset of features is selected for consideration at *every node split* during tree construction, leading to greater randomness.

Feature Sampling Mechanism
🌲 In Bagging, if 2 out of 5 columns are sampled for a tree, that tree only uses those 2 columns for all its splits (e.g., using only $x_0$ and $x_1$).
🌲 In Random Forest, if 2 features are sampled, these features are re-sampled randomly at each node. For example, one node might split on $\{x_1, x_3\}$, and the next node might split on $\{x_4, x_5\}$.
🌲 This node-level sampling in Random Forest increases the randomness and diversity among the base estimators, which generally leads to better performance compared to standard Bagging.

Key Points & Insights
➡️ A key interview differentiator is that Bagging is a general ensemble technique, whereas Random Forest imposes the constraint of using Decision Trees exclusively.
➡️ If a Bagging Classifier is constructed using only Decision Trees, the distinction remains in the feature sampling method (tree-level vs. node-level).
➡️ Random Forest's node-level feature selection (introducing extra randomness) is the reason it often outperforms Bagging models in practical scenarios.
➡️ Always be prepared to explain the counter-question: If Bagging uses only Decision Trees, why isn't it a Random Forest? The answer is the feature sampling strategy.

📸 Video summarized with SummaryTube.com on Nov 27, 2025, 10:04 UTC