Variable priority for unsupervised variable selection

Lili Zhou

doi:10.1016/j.patcog.2025.112727

Variable priority for unsupervised variable selection

2025-11-15

0

PRE

AI

DOI：10.1016/j.patcog.2025.112727

Original

Original request for help

Save

Abstract

En 中文

• Introduces a lasso-based framework for unsupervised feature selection using proxy classification. • Leverages decision trees to create local rules that reveal which variables drive structure. • Converts unlabeled data into supervised tasks, making hidden signals easier to detect. • Averages across many trees to reduce variance and produce stable, interpretable importance scores. • Shows strong performance on synthetic, biological, and image datasets compared with leading methods.

Variable priority for unsupervised variable selection

Abstract

Journal

Organization