A novel game-based belief rule base

Abstract

Establishing an initial belief rule base (BRB) expert system typically relies on expert knowledge. However, the heterogeneity in knowledge arising from variations in expert cognition directly impacts the accuracy and robustness of BRB. Furthermore, the existing research lacks a discussion on building the initial BRB under multi-expert knowledge. In response to this challenge, a novel game-based belief rule base (GBRB) is proposed in this paper. First, a consensus matrix is introduced to formalize the interaction among expert knowledge, which is the foundation for constructing a multi-expert evolutionary game. Subsequently, the replicator equation for population dynamics is employed to simulate the dynamic process of the multi-expert evolutionary game. Finally, optimal expert knowledge, which refers to referential values and belief degrees, is determined based on the evolution results to participate in establishing GBRB. Case studies are conducted to illustrate the construction process and effectiveness of the GBRB. The experimental results are compared with machine learning-related methodologies. The comparative results demonstrate the good performance of GBRB in modeling effectiveness and prediction accuracy.