Università degli Studi di Cagliari
---
Nell’ambito dei seminari di Statistica, mercoledì 10 giugno alle ore 11:30 presso l’aula B, Andrea Carta, post-doc di Statistica al Dipartimento di Scienze Economiche e Aziendali, terrà il seguente seminario: "Oblique Decision Trees and Oblique Random Forests for Regression via Weighted Support Vector Machines"
Abstract: Standard decision trees partition the space of variables using axis-aligned splits, which can be suboptimal in the presence of correlated predictors or complex data geometries. This talk presents a unified framework for learning oblique decision trees and ensembles for regression, based on the combination of correlation-based variable selection and weighted linear support vector machines. At each internal node, the splitting hyperplane is identified by applying a weighted linear support vector machine to a binarized response, after screening predictors by their correlation with the outcome. This strategy naturally extends to the ensemble setting, where oblique trees serve as base learners within a random forest framework, complemented by a variable importance measure tailored to oblique splits. Experiments on simulated and benchmark datasets show that both the single-tree and ensemble methods outperform their axis-aligned counterparts, with particularly robust gains in the presence of continuous predictors and complex interactions among variables.
Il seminario potrà essere seguito anche su Microsoft Teams al link: