Approximate Reduction in IntervalValued Ordered Target Information Systems Based on Attribute Weighting

Abstract

Attribute reduction is a critical preprocessing step in interval-valued ordered target information systems. However, conventional methods often inadvertently discard decision critical attributes during dimensionality reduction. This oversight severely degrades subsequent classification and decision making performance. To mitigate this issue, this paper develops a novel attribute weighted reduction framework that preserves essential attributes while compressing redundant features. We first formally introduce interval-valued ordered target information systems with attribute importance, and systematically establish the theoretical foundation of upper and lower approximation reductions. To facilitate efficient computation, we convert interval data into scalar representations by combining interval maxima and radii, constructing a transformed ordered target system that retains both numerical magnitude and inherent uncertainty. On this basis, we design a weighted discernibility matrix to characterize attribute distinguishability under varying importance levels. We further derive rigorous necessary and sufficient conditions for identifying upper and lower approximation reducts, enabling reliable reduction with explicit consideration of attribute weights. The rationality and correctness of the theoretical results are verified through illustrative examples. Extensive experiments on multiple real world datasets demonstrate that the proposed method achieves effective dimensionality reduction. It also consistently retains decision critical attributes.OPEN ACCESS Received: 02/04/2026 Accepted: 02/06/2026 Published: 21/07/2026

Document