Zobraziť v EDS

Multilabel feature selection using graph neural networks and differential evolution optimization.

Uložené v:
Podrobná bibliografia
Názov: Multilabel feature selection using graph neural networks and differential evolution optimization.
Autori: Pan, Ning
Zdroj: Scientific Reports; 12/23/2025, Vol. 15 Issue 1, p1-22, 22p
Predmety: GRAPH neural networks, DIFFERENTIAL evolution, FEATURE selection, MACHINE learning, DATA structures, MATHEMATICAL optimization, DIMENSIONAL reduction algorithms
Abstrakt: In recent years, the exponential growth of high-dimensional data across domains such as data mining, machine learning, bioinformatics, social network analysis, and image processing has amplified the importance of feature selection as a primary method for dimensionality reduction. High-dimensional datasets often include redundant, noisy, and irrelevant features that hinder the development of accurate, efficient, and reliable learning models. This issue becomes even more critical in multi-label scenarios, where each instance may be associated with multiple labels, creating intricate interactions between features and labels as well as interdependencies among the labels themselves. Traditional feature selection techniques, designed primarily for single-label problems, struggle to address the complexities of multi-label data. Consequently, there is a pressing need for novel approaches that can effectively capture and utilize these complexities. This study introduces a hybrid method that combines the representational power of Graph Neural Networks (GNNs) with the optimization efficiency of Differential Evolution (DE) to tackle multi-label feature selection challenges. GNNs leverage graph structures to model complex relationships between features and labels, allowing for simultaneous consideration of feature relevance and label dependencies. DE, a robust global optimization algorithm, ensures the selection of an optimal subset of features by exploring vast search spaces and avoiding local optima. The proposed method was evaluated on six diverse datasets spanning text and image domains. GNN-DE achieves optimal classification performance while selecting fewer features, demonstrating its efficiency in reducing computational complexity. For datasets with complex label correlations, such as Enron and Scene, this method outperforms existing approaches. [ABSTRACT FROM AUTHOR]
Copyright of Scientific Reports is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Databáza: Complementary Index
Popis
Abstrakt:In recent years, the exponential growth of high-dimensional data across domains such as data mining, machine learning, bioinformatics, social network analysis, and image processing has amplified the importance of feature selection as a primary method for dimensionality reduction. High-dimensional datasets often include redundant, noisy, and irrelevant features that hinder the development of accurate, efficient, and reliable learning models. This issue becomes even more critical in multi-label scenarios, where each instance may be associated with multiple labels, creating intricate interactions between features and labels as well as interdependencies among the labels themselves. Traditional feature selection techniques, designed primarily for single-label problems, struggle to address the complexities of multi-label data. Consequently, there is a pressing need for novel approaches that can effectively capture and utilize these complexities. This study introduces a hybrid method that combines the representational power of Graph Neural Networks (GNNs) with the optimization efficiency of Differential Evolution (DE) to tackle multi-label feature selection challenges. GNNs leverage graph structures to model complex relationships between features and labels, allowing for simultaneous consideration of feature relevance and label dependencies. DE, a robust global optimization algorithm, ensures the selection of an optimal subset of features by exploring vast search spaces and avoiding local optima. The proposed method was evaluated on six diverse datasets spanning text and image domains. GNN-DE achieves optimal classification performance while selecting fewer features, demonstrating its efficiency in reducing computational complexity. For datasets with complex label correlations, such as Enron and Scene, this method outperforms existing approaches. [ABSTRACT FROM AUTHOR]
ISSN:20452322
DOI:10.1038/s41598-025-27824-x