ABSTRACT
A decision tree is a non-parametric supervised learning tool for extracting knowledge from available data. It is non-parametric, since it does not require any assumptions regarding the distribution of the dependent variable, the explanatory variables, and the functional form of the relationships between them. Decision trees are discriminated according to the nature of the target they have to predict. A classification tree is characterized by the fact that it predicts a categorical response, as opposed to a quantitative and, generally, continuous one in the regression case. The rationale behind any split, at any internal node, is to generate two descendant child nodes where the data are increasingly pure. A simple visual representation might clarify the concept. The choice of the size is a fundamental step in the tree building process. The problem of lack of independence between training set and test set suggests to formulate an alternative known as test-sample estimate.
