Statistical Advantages of Oblique Randomized Decision Trees and Forests

The article titled "Statistical Advantages of Oblique Randomized Decision Trees and Forests," published on arXiv in November 2025, investigates the benefits of incorporating features formed by general linear combinations of covariates within randomized decision tree and forest regression algorithms. This approach enables oblique splits, which differ from traditional axis-aligned splits by allowing decision boundaries at arbitrary angles. The study provides a theoretical analysis of a novel class of random tree and forest estimators that utilize these oblique splits, drawing on concepts from random tessellation theory in stochastic geometry. By grounding the analysis in this theoretical framework, the research offers insights into the statistical properties and potential advantages of oblique randomized decision trees and forests. This work contributes to the broader field of machine learning algorithms by expanding the understanding of how feature construction and split strategies impact model performance. The article aligns with recent developments in algorithmic theory and stochastic geometry, as reflected in connected research on similar topics. Overall, the study advances knowledge on the design and analysis of decision tree-based regression methods using oblique partitioning techniques.