<p>The prevalence of multi-label (ML) data has witnessed a significant increase in numerous fields as big data technology continues to expand. However, these datasets often contain a high degree of redundancy and irrelevant attributes, which can negatively impact the efficiency and predictive performance of machine learning models. To address these challenges, this paper introduces PSI-MFS, a novel lightweight multi-objective feature selection (MLFS) approach that efficiently optimizes feature selection criteria while maintaining computational efficiency. Despite the availability of several MLFS approaches, many existing methods struggle to achieve optimal balance between computational efficiency and selection quality. PSI-MFS overcomes this by simultaneously optimizing three conflicting feature selection (FS) objectives: minimizing feature–feature redundancy (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7163_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="14" /> </InlineMediaObject> <EquationSource Format="TEX">$\downarrow $</EquationSource> <EquationSource Format="MATHML"><math> <mo stretchy="false">↓</mo> </math></EquationSource> </InlineEquation>), maximizing feature–label relevancy (<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7163_Article_IEq2.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="14" /> </InlineMediaObject> <EquationSource Format="TEX">$\uparrow $</EquationSource> <EquationSource Format="MATHML"><math> <mo stretchy="false">↑</mo> </math></EquationSource> </InlineEquation>), and maximizing feature–label interaction (<InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7163_Article_IEq3.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="14" /> </InlineMediaObject> <EquationSource Format="TEX">$\uparrow $</EquationSource> <EquationSource Format="MATHML"><math> <mo stretchy="false">↑</mo> </math></EquationSource> </InlineEquation>), using a preference selection index (PSI)-based optimization strategy. PSI-MFS provides a trade-off between feature correlation and classification performance while significantly reducing memory consumption and execution time. The time complexity of PSI-MFS is low, and experimental assessments on ten benchmark datasets demonstrate its superior or competitive performance compared to 11 state-of-the-art (SOTA) FS methods. The effectiveness of PSI-MFS is further validated through statistical analysis using Friedman’s test, which confirms its significant performance improvements. Notably, PSI-MFS achieves faster execution times and outperforms existing methods in 80% of cases, making it a robust and scalable solution for ML feature selection.</p>

PSI-MFS: lightweight multi-objective feature selection for enhanced multi-label classification

期刊论文
发表日期：

2025年4月18日

查看全文

Gurudatta Verma,
Tirath Prasad Sahu

摘要

The prevalence of multi-label (ML) data has witnessed a significant increase in numerous fields as big data technology continues to expand. However, these datasets often contain a high degree of redundancy and irrelevant attributes, which can negatively impact the efficiency and predictive performance of machine learning models. To address these challenges, this paper introduces PSI-MFS, a novel lightweight multi-objective feature selection (MLFS) approach that efficiently optimizes feature selection criteria while maintaining computational efficiency. Despite the availability of several MLFS approaches, many existing methods struggle to achieve optimal balance between computational efficiency and selection quality. PSI-MFS overcomes this by simultaneously optimizing three conflicting feature selection (FS) objectives: minimizing feature–feature redundancy ( $\downarrow $ $↓$ ), maximizing feature–label relevancy ( $\uparrow $ $↑$ ), and maximizing feature–label interaction ( $\uparrow $ $↑$ ), using a preference selection index (PSI)-based optimization strategy. PSI-MFS provides a trade-off between feature correlation and classification performance while significantly reducing memory consumption and execution time. The time complexity of PSI-MFS is low, and experimental assessments on ten benchmark datasets demonstrate its superior or competitive performance compared to 11 state-of-the-art (SOTA) FS methods. The effectiveness of PSI-MFS is further validated through statistical analysis using Friedman’s test, which confirms its significant performance improvements. Notably, PSI-MFS achieves faster execution times and outperforms existing methods in 80% of cases, making it a robust and scalable solution for ML feature selection.