Analysis of the generalization ability of a full decision tree

Classification algorithms based on full decision trees are investigated. Due to the decision tree construction under consideration, all the features satisfying a branching criterion are taken into account at each special vertex. The generalization ability of a full decision tree is estimated by applying margin theory. It is shown on real-life problems that the construction of a full decision tree leads to an increase in the margins of the learning objects; moreover, the number of objects with a positive margin increases as well. It is shown that the empirical Rademacher complexity of a full decision tree is lower than that of a classical decision tree.