Nonlinear Optimal Classification Trees

This paper presents nonlinear optimal classification trees, an advanced approach developed to enhance the accuracy and interpretability of classification trees. By integrating kernel-based methods and symbolic functions, our models provide a more precise approximation of decision boundaries, resulting in compact and interpretable trees. Additionally, we introduce an improved margin-shifting algorithm that increases the model's robustness. Experiments on 21 public datasets confirm that our methods can significantly surpass traditional decision trees and previous optimal tree models in and with more compact models, establishing them as an effective solution for complex classification tasks.