Random Forests

Random Forests (RF), introduced by Breiman (Machine Learning, 45(1):5–32, 2001), are a nonparametric statistical method dealing with classification and regression problems. RF is one of the main methods used in machine learning to tackle prediction problem. Their excellent predictive performance as well as their flexibility, imposing mild restrictions on the nature of the data, lead to intensive use in various fields of application. Indeed, they are adapted to both supervised classification and regression problems and allow to consider together nominal and numerical explanatory variables without pre-processing. They are now one of the favorite methods in the toolbox of statisticians. Let us start this chapter by describing RF in the classical regression framework, without any reference to the time series context. Since forests are made of trees, we also briefly mention the CART (Classification and Regression Trees) algorithm.