# Rolling Lookahead Learning for Optimal Classification Trees

arXiv: https://arxiv.org/abs/2304.10830 · HTML: https://mlsanity.com/paper/2304.10830/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 8 of its 49 references that are in the map.
- First posted on arXiv 21 Apr 2023.

## Builds on (most cited first, 8 of 8)

- [Optimal Sparse Decision Trees](https://mlsanity.com/paper/1904.12847/) (2019)
- [Generalized and Scalable Optimal Sparse Decision Trees](https://mlsanity.com/paper/2006.08690/) (2020)
- [Learning Optimal and Fair Decision Trees for Non-Discriminative Decision-Making](https://mlsanity.com/paper/1903.10598/) (2019)
- [Strong Optimal Classification Trees](https://mlsanity.com/paper/2103.15965/) (2021)
- [Computing Optimal Decision Sets with SAT](https://mlsanity.com/paper/2007.15140/) (2020)
- [Optimal randomized classification trees](https://mlsanity.com/paper/2110.11952/) (2021)
- [Sparsity in Optimal Randomized Classification Trees](https://mlsanity.com/paper/2002.09191/) (2020)
- [Learning Optimal Classification Trees: Strong Max-Flow Formulations](https://mlsanity.com/paper/2002.09142/) (2020)

## Built on by (most cited first, 1 of 1)

- [An Exact Junction-Tree Extended Formulation for Optimal Classification Trees](https://mlsanity.com/paper/2609.24741/) (2026)

## Abstract (arXiv)

Classification trees continue to be widely adopted in machine learning applications due to their inherently interpretable nature and scalability. We propose a rolling subtree lookahead algorithm that combines the relative scalability of the myopic approaches with the foresight of the optimal approaches in constructing trees. The limited foresight embedded in our algorithm mitigates the learning pathology observed in optimal approaches. At the heart of our algorithm lies a novel two-depth optimal binary classification tree formulation flexible to handle any loss function. We show that the feasible region of this formulation is an integral polyhedron, yielding the LP relaxation solution optimal. Through extensive computational analyses, we demonstrate that our approach outperforms optimal and myopic approaches in 808 out of 1330 problem instances, improving the out-of-sample accuracy by up to 23.6% and 14.4%, respectively.