Sparse Decision Tree

Sparse decision trees are interpretable machine learning models aiming to achieve high predictive accuracy with minimal complexity. Current research focuses on developing provably optimal algorithms, often employing dynamic programming and branch-and-bound techniques, to construct these trees efficiently, even for large datasets and weighted samples. This work addresses challenges like covariate shift and the exploration of the "Rashomon set" – the multitude of near-optimal models – to improve model selection and enhance explainability, particularly in high-stakes applications such as healthcare and policy design. The resulting improvements in both accuracy and interpretability are significant for various fields requiring transparent and reliable decision-making.