A novel multi-criteria conflict evidence combination method and its application to pattern recognition

Abstract

In recent years, the Dempster–Shafer Theory (DST) has been widely applied in areas such as target classification and multi-modal fusion due to its advantages in uncertain reasoning. However, in DST, when there exists highly conflicts between Sources of Evidence (SoEs), it often leads to counterintuitive fusion results, thereby affecting the performance of the final fused decision-making. To eliminate the potential impact of highly conflicts during the fusion process, this paper primarily focuses on modifying the SoEs themselves to achieve the discounting fusion. Currently, existing literature has proposed numerous discounting fusion methods to address highly conflict fusion. However, in these discounting strategies, the determination of discounting factors is typically based on a single criterion. Considering that a single indicator cannot comprehensively and accurately assess the reliability of each evidence source, this paper introduces a novel conflict evidence combination method based on a multi-criteria evaluation strategy. In this method, the Best and Worst Method (BWM) is initially used to prioritize the best SoE, determining the solution points for each criterion and subsequently selecting the ideal solution. Then, combining the selected criteria: distance, relative entropy, and divergence metrics, reliability calculation is performed based on the Stable Preference Order Theory for Ideal Solution (SPOTIS), which achieves a stable preference order strategy towards the ideal solutions. Finally, leveraging the discounting fusion method in DST, modification and fusion of the original highly conflict SoEs are achieved. Through extensive experiments, the effectiveness and practicality of the proposed method in this paper have been validated.