Interpretable Optimal Classification Tree for the Stress Level Classification Problem

AI-based solutions have been widely adopted, but they are not always appropriate when interpretability is essential. Recent models can generate more accurate classification trees than heuristic-based ones by using optimization. We trained the Classification and Regression Tree (CART) and the Optimal Classification Tree using Mixed-Integer Optimization (OCT-MIO) models using cross-validation on dataset of people classified into 5 stress levels. The experiments show that the OCT-MIO model achieves more correct predictions than CART for the same tree depth, making it a better alternative for problems where higher accuracy is required while preserving interpretability.