Particle swarm optimisation for sparsity-based feature selection in multi-label classification
Lists
Only you see this
Abstract
Multi-label classification (MLC) is an emerging real-world problem in which each instance is associated with multiple class labels simultaneously. Multi-label classification is challenging due to the complex interactions of features and multiple labels. Sparsity-based feature selection is an effective and efficient approach to selecting relevant features for multi-label classification. However, most (if not all) sparsity-based approaches are gradient-based and thus they tend to get stuck at local optima. This paper proposes a new sparsity approach based on particle swarm optimisation (PSO) which enhances its global search ability, thereby avoiding local optima. The paper also proposes new sparsity-based fitness functions for PSO, which can consider the feature interactions. The experimental results show that PSO can enable sparsity-based methods to select highly relevant features, thus improving the multi-label classification performance. The proposed sparsity-based methods also achieve the highest number of statistically significant results in comparison to several state-of-the-art and standard sparsity-based benchmarks.
BibTeX
@inproceedings{demir2022particle,
author = {Kaan Demir and Bach Hoai Nguyen and Bing Xue and Mengjie Zhang},
title = {{Particle swarm optimisation for sparsity-based feature selection in multi-label classification}},
booktitle = {GECCO Companion},
year = {2022},
doi = {10.1145/3520304.3529074},
}Loading related content...