Classification and Regression Trees (CART)

Splits data based on binary rules. The model predicts based on which region data falls under. Regions can only be split horizontally or vertically.

To build the tree, use a method called Recursive Binary Splitting

At a high level, we would split the predictor space $(X_{1}, X_{2}, ..., X_{n})$ into $J$ non-overlapping regions $R_{1}, R_{2}, .. R_{J}$ .

In a Regression problem, we would average the points of region $R_{i}$ for $i \in 1, 2, ..., J$ .

In a classification problem, we would take the majority vote of the points in region $R_{i}$ .

Quartz 4