Optimal Classification Tree

Optimal classification trees aim to construct decision trees that maximize classification accuracy while minimizing tree complexity, often using mixed-integer programming (MIP) formulations. Current research focuses on developing efficient MIP models incorporating various loss functions (e.g., logistic loss), robustness to data distribution shifts, and fairness constraints, often leveraging techniques like Boolean rules or support vector machine (SVM) based splits. These advancements enhance the interpretability and predictive power of classification trees, impacting fields requiring both high accuracy and model explainability, such as healthcare and social sciences.