Classification and Regression Trees (CART) are a type of decision tree algorithm used for predicting outcomes by learning decision rules from data features. They work by recursively splitting data into subsets based on feature values, aiming to maximize the separation of classes or minimize prediction error in regression tasks.