In the previous section, we built multiple models for the same classification problem. The bootstrapped trees were generated by using resamples of the observations. Breiman (2001) suggested an important variation—actually, there is more to it than just a variation—where a CART is built with the covariates (features) being resampled for each of the bootstrap samples of the dataset. Since the final tree of each bootstrap sample has different covariates, the ensemble of the collective trees is called a Random Forest. A formal algorithm is given next.