Generalized Ordinal Priority Approach for Multi-Attribute Decision-Making under Incomplete Preference Information

TL;DR

This paper introduces a generalized MADM method called GOPA that effectively handles incomplete preference information by combining utility distribution estimation and optimization, improving decision reliability and analytical solvability.

Contribution

It develops a new framework for MADM under incomplete preferences, establishing properties of the approach and providing a practical solution method with validation metrics.

Findings

GOPA offers model generalizability and analytical solutions.

It effectively handles incomplete preference data.

The method is validated through an emergency supplier selection case.

Abstract

The Ordinal Priority Approach (OPA) is a multi-attribute decision-making (MADM) method to determine the relative importance (weights) of experts, attributes, and alternatives. This study formally establishes the fundamental properties of OPA, including solution efficiency, analytical solution expression, the decomposability of optimal decision weights, and its relationship with rank-based surrogate weights. Building on these properties, we propose a Generalized Ordinal Priority Approach (GOPA) based on an "estimate-then-optimize" contextual optimization framework for MADM when preference information is incomplete. In the first stage, we derive utility distributions for ranked alternatives in discrete and continuous prospects by minimizing cross-entropy utility under partial preference information, including weak order relations, absolute differences, ratio scales, and lower bounds. Rank-based surrogate weights and risk preference utility functions serve as the global utility structure for discrete and continuous prospects, respectively. The elicited utility information is then introduced into the second-stage problem to simultaneously optimize the weights of experts, attributes, and alternatives within a normalized weight space. Metrics for validating the group decision outcomes of GOPA, including percentage standard deviation, correlation coefficient, and confidence level measurement, are proposed. Theoretical analysis reveals several advantageous properties of GOPA, including model generalizability, analytical solvability, and risk preference independence. Furthermore, this study provides a lower bound reference for transforming the general optimization-based weight elicitation problems into optimization problems with stochastic dominance constraints. The applicability of GOPA is demonstrated through an improvisational emergency supplier selection problem.