Sparse Probabilistic Splits for Tree Ensembles: A Tunable Regularizer for Stable Generalization on Tabular Data

This paper proposes a regularization technique called sparse probabilistic split (SPS), which introduces controlled randomness into the node-splitting process of decision trees to mitigate overfitting and enhance generalization. Building on this technique, we propose two models: the sparse probabili...

Full description

Bibliographic Details

Main Authors:	Jongkwan Choi, Yujin Lee, Eunbin Yun, Gyeongtaek Lee
Format:	Artigo
Language:	Inglês
Published:	IEEE 2026-01-01
Series:	IEEE Access
Subjects:	Decision trees sparse probabilistic split ensemble learning model generalization random splitting
Online Access:	https://ieeexplore.ieee.org/document/11339487/
Tags:	Add Tag No Tags, Be the first to tag this record!