Partial-overall dominance three-way decision models in interval-valued decision systems

Abstract

Three-way decisions are a generalization of classical decision theory and receive increasing attentions from various fields to handle decision-making problems, especially when involving in incomplete information. An interval is a typical notion of information representation with incompleteness and uncertainty. To measure the dominance degree of one interval dominating or being dominated by another is a hot issue. In this paper, a novel dominance measure is constituted by considering both lengths and locations of intervals. The proposed dominance measure distinguishes two overlapped or coincided intervals, and is able to quantize the separation degree of disjoint intervals. A notion of variable precision overall dominance relation is introduced by integrating both the attribute-wise evaluation information and overall dominance degree of objects. Based on the constituted dominance relation, two three-way decision models are presented in interval-valued decision systems with categorical and interval-valued decision attributes, respectively. Numerical examples of three-way decisions and simulated experiments on classification over two synthetic and four benchmark datasets are presented. Experimental results indicate that the proposed model can produce lower classification errors in comparison with existing methods.