Comprehensive nonmodularity and interaction indices for decision analysis

Article type: Research Article

Authors: Wu, Jian-Zhang^{a; b; *} | Beliakov, Gleb^b

Affiliations: [a] Business School, Ningbo University, Ningbo, China | [b] School of Information Technology, Deakin University, Burwood, Australia

Correspondence: [*] Corresponding author. Jian-Zhang Wu. E-mail: [email protected].

Abstract: Nonmodularity is a prominent property of capacity that deeply links to the internal interaction phenomenon of multiple decision criteria. Following the common architectures of the simultaneous interaction indices as well as of the bipartition interaction indices, in this paper, we construct and study the notion of probabilistic nonmodularity index and also its particular cases, such as Shapely and Banzhaf nonmodularity indices, which can be used to describe the comprehensive interaction situations of decision criteria. The connections and differences among three categories of interaction indices are also investigated and compared theoretically and empirically. It is shown that three types of interaction indices have the same roots in their first and second orders, but meanwhile the nonmodularity indices have involved less amount of subsets and can be adopted to describe the interaction phenomenon in decision analysis.

Keywords: Fuzzy measure, interaction representation, nonmodularity, nonadditivity, decision analysis

DOI: 10.3233/JIFS-201583

Journal: Journal of Intelligent & Fuzzy Systems, vol. 40, no. 6, pp. 10671-10685, 2021