Árboles de decisión y Random Forests

A decision tree classifies by asking a chain of yes-or-no questions about your features until it reaches an answer. 🌳

Every branch point is a split, a rule like is age above 30. The data flows left or right based on that test.

When no more useful questions remain, you hit a leaf. The class that dominates that leaf becomes the prediction.

A tree chooses each split to make groups as pure as possible, measured by criteria like Gini impurity or entropy.

One real strength is interpretability: you can follow the path of questions and explain exactly why a prediction was made.

Left unchecked, a tree grows deep and memorizes the training data. That overfitting hurts accuracy on new, unseen rows.

You tame a tree by capping its growth with max_depth, trading a little training fit for far better generalization.

A random forest trains many trees and lets them vote. The crowd is far steadier than any single deep tree.

For variety, each tree trains on a random sample of rows and a random subset of features. That diversity is the secret sauce.

It is one line in scikit-learn. Set n_estimators to choose how many trees join the ensemble.

A trained forest tells you which inputs mattered most through feature_importances_, a quick guide to what drives predictions.

Let's confirm why a forest improves on a single tree.

A decision tree asks questions down to a leaf; a random forest votes across many trees to cut overfitting. 🎯