Rotation-based model trees for classification

Structurally, a model tree is a regression method that takes the form of a decision tree with linear regression functions instead of terminal class values at its leaves. In this study, model trees were coupled with a rotation-based ensemble for solving classification problems. In order to apply this regression technique to classification problems, we considered the conditional class probability function and sought a model-tree approximation to it. During classification, the class whose model tree generated the greatest approximated probability value was chosen as the predicted class. We performed a comparison with other well-known ensembles of decision trees on standard benchmark data sets, and the performance of the proposed technique was greater in most cases.