What Is Cart?

CART means classification and regression tree. It is a non-parametric approach for developing a predictive model. What is meant by non-parametric is that in implementing this methodology, we do not have any assumption that our target variable has to follow certain probability distribution. This is a big relaxation from developing parametric models (for example linear regression models etc.) where we need to check the model's adequacy by assessing the various underlying assumption. When our target variable is categorical, then we say it is a classification problem and when our target variable is continuous i.e quantitative by nature then we call it is a regression problem.

Combination of classification and regression problem together is termed as CART. CART models are supervised models in terms of Machine Learning literature. These type of models are basically tree based (classification or regression) models. In developing such models, each node is basically splitted in binary fashion (not more than 2 split). For regression tree, target value of any new observation is estimated by averaging out the value of the training set observations that followed a particular branch where as in classification problem the predicted class is determined with certain probability.