Self-adaptive interval dominance-based feature selection for monotonic classification of interval-valued attributes

Dominance rough set theory is a key mathematical tool for addressing monotonic classification tasks (MCTs). However, current dominance rough set models for feature selection are highly sensitive to noise. Furthermore, in practical applications, conditional feature attributes are often expressed as interval values. The endpoints of these intervals are susceptible to noise pollution, resulting in incomplete knowledge and interference with feature selection. To address these issues, this paper proposes a feature selection approach named SIDDM-FDNRS with three distinctive characteristics: (i) An interval-valued fuzzy dominance neighborhood rough set (IV-FDNRS) model is established to reduce noise interference by introducing adaptive neighborhood parameters. (ii) Definition of a novel self-adaptive interval dominance discernibility matrix (SIDDM) based on the IV-FDNRS model, used for selecting the optimal monotonic features. (iii) Comprehensive consideration of interval dominance relationships, the strength of global dominance relationships, and noise interference factors in the proposed feature selection model. Extensive experiments demonstrate that the proposed feature selection algorithm leads to superior classification performance.