Fast multilabel feature selection via global relevance and redundancy optimization
Contribution
Selects multilabel features by jointly optimizing relevance and redundancy, fast enough for practice.
Abstract
Multi-label feature selection faces a fundamental trade-off: features relevant for one label may be irrelevant or noisy for others, and naive global ranking discards useful label-specific structure. This paper proposes a fast multilabel feature selection framework that jointly optimizes global relevance and redundancy through a global optimization objective, achieving state-of-the-art accuracy across diverse multi-label benchmarks while running orders of magnitude faster than competing methods.
Why This Paper Matters
A practical, fast drop-in replacement for multi-label feature selection that improves results on standard benchmarks. Useful for anyone working on multi-label classification, tag recommendation, or hierarchical text categorization.
BibTeX
@article{zhang2022fast,
title={Fast multilabel feature selection via global relevance and redundancy optimization},
author={Zhang, Jia and Lin, Yidong and Jiang, Min and Li, Shaozi and Tang, Yong and Long, Jinyi and Weng, Jian and Tan, Kay Chen},
journal={IEEE Transactions on Neural Networks and Learning Systems},
volume={35},
number={4},
pages={5721--5734},
year={2022},
publisher={IEEE},
}