Machine Learning Classic: Parsimonious Binary Classification Trees
Get your hands on a classic technical report outlining a three-step method of construction binary decision trees for multiple classification problems.
By Leo Breiman and Charles J. Stone.
A three-step method of construction binary decision trees for multiple classification problems is presented. First a splitting rule is defined in terms of a generalization of Gini’s index of diversity. Next the optimal termination rule is found relative to a criterion which penalizes both misclassifications and complex trees (i.e., those having many terminal nodes. The tree thus obtained depends on a complexity parameter which, in the final step is selected by data-splitting or cross-validation.
