A Moving-Horizon Approximate Branch-and-Reduce Method for Deep Classification Trees
A new moving-horizon branch-and-reduce method enables training near-optimal deep classification trees on large-scale datasets with continuous features.
Researchers have introduced a hierarchical root-subtree optimization framework to address the scalability issues of deep decision trees. By combining branch-and-reduce for root-level problems with approximation techniques, the method overcomes traditional limitations in feature selection and tree depth, offering a more efficient path to interpretable, high-accuracy models.