Cost-sensitive feature selection for multi-label classification: multi-criteria decision-making approach

Purpose In multi-label classification, selecting the most relevant features is crucial for enhancing predictive performance and reducing computational complexity. Real-world scenarios often involve significant costs in data acquisition, including time, financial and computational resources. However, most existing feature selection methods overlook the associated costs. Design/methodology/approach Multicriteria decision-making (MCDM) has emerged as a powerful tool for addressing complex problems involving multiple, often conflicting criteria. This study proposes a novel cost-sensitive multi-label feature selection method that fuses feature importance with feature cost within an MCDM framework. The proposed method transforms a cost-sensitive multi-label feature selection problem into an MCDM problem by leveraging mutual information. Furthermore, the data were converted into Fermatean fuzzy sets, and the Fermatean fuzzy simple weighted sum product (WISP) method was employed to rank features based on their relevance to labels and associated costs. Findings Extensive experiments conducted on ten benchmark datasets against five evaluation metrics demonstrated the superiority of the proposed method in selecting relevant features while minimizing costs and consistently outperforming existing methods. Originality/value Unlike existing methods that integrate costs through penalties and select features via a greedy search, the proposed approach adopts an MCDM-based strategy for feature ranking. This method aims to achieve globally optimal outcomes by balancing the trade-offs between conflicting objectives, marking a significant advancement over existing techniques. Graphicalabstract