How it works...

A random forest is a special regressor formed of a set of simple regressors (decision trees), represented as independent and identically distributed random vectors, where each one chooses the mean prediction of the individual trees. This type of structure has made significant improvements in regression accuracy and falls within the sphere of ensemble learning. Each tree within a random forest is constructed and trained from a random subset of the data in the training set. The trees therefore do not use the complete set, and the best attribute is no longer selected for each node, but the best attribute is selected from a set of randomly selected attributes.

Randomness is a factor that then becomes part of the construction of regressors and aims to increase their diversity and thus reduce correlation. The final result returned by the random forest is nothing but the average of the numerical result returned by the different trees in the case of a regression, or the class returned by the largest number of trees if the random forest algorithm was used to perform classification.