Multi-source multi-label feature selection with missing features

Yabo Shi

doi:10.1016/j.eswa.2025.129879

Multi-source multi-label feature selection with missing features

2025-10-04

0

PRE

AI

DOI：10.1016/j.eswa.2025.129879

Original

Original request for help

Save

Abstract

En 中文

• Dual-problem joint modeling: A unified framework addressing both feature missing and label skewness in multi-source multi-label (MSML) data. Missing features are dynamically imputed via a feature correlation matrix, while multi-label oversampling mitigates label imbalance. • Enhanced feature selection algorithm: The classic Infinite Feature Selection (Inf-FS) is improved by integrating label auxiliary information and label-specific feature correlations, significantly boosting accuracy in multi-label scenarios. • Cross-source fusion strategy: A redundancy-aware feature integration framework replaces conventional intersection/union methods, ensuring globally optimal feature subsets through inter-source redundancy analysis.

Multi-source multi-label feature selection with missing features

Abstract

Journal

Organization

Cited Papers