Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Guaranteeing consistency in evidence fusion: A novel perspective on credibility

Chaoxiong Ma, Yan Liang, Huixia Zhang, Hao Sun

TL;DR

The paper tackles a core problem in evidence fusion: credibility and fusion results are often computed in separate, open-loop steps, leading to inconsistencies. It introduces Iterative Credible Evidence Fusion ($ICEF$), a closed-loop framework that feeds the fusion output back into credibility calculation, and a novel plausibility-belief arithmetic-geometric divergence ($PBAGD$) to quantify and downweight conflicting evidence. The approach defines conditional credibility and jointly optimizes event probabilities, credibility, and fusion results, using an event-evaluation mechanism and an iterative update rule based on $BetP$ posterior probabilities. Across numerical simulations and five UCI benchmarks, ICEF–PBAGD demonstrates improved consistency, higher decision-support quality, and robustness to outlier/conflicting sources, highlighting its practical value for credible multi-source fusion.

Abstract

It is explored that available credible evidence fusion schemes suffer from the potential inconsistency because credibility calculation and Dempster's combination rule-based fusion are sequentially performed in an open-loop style. This paper constructs evidence credibility from the perspective of the degree of support for events within the framework of discrimination (FOD) and proposes an iterative credible evidence fusion (ICEF) to overcome the inconsistency in view of close-loop control. On one hand, the ICEF introduces the fusion result into credibility assessment to establish the correlation between credibility and the fusion result. On the other hand, arithmetic-geometric divergence is promoted based on the exponential normalization of plausibility and belief functions to measure evidence conflict, called plausibility-belief arithmetic-geometric divergence (PBAGD), which is superior in capturing the correlation and difference of FOD subsets, identifying abnormal sources, and reducing their fusion weights. The ICEF is compared with traditional methods by combining different evidence difference measure forms via numerical examples to verify its performance. Simulations on numerical examples and benchmark datasets reflect the adaptability of PBAGD to the proposed fusion strategy.

Guaranteeing consistency in evidence fusion: A novel perspective on credibility

TL;DR

The paper tackles a core problem in evidence fusion: credibility and fusion results are often computed in separate, open-loop steps, leading to inconsistencies. It introduces Iterative Credible Evidence Fusion (), a closed-loop framework that feeds the fusion output back into credibility calculation, and a novel plausibility-belief arithmetic-geometric divergence () to quantify and downweight conflicting evidence. The approach defines conditional credibility and jointly optimizes event probabilities, credibility, and fusion results, using an event-evaluation mechanism and an iterative update rule based on posterior probabilities. Across numerical simulations and five UCI benchmarks, ICEF–PBAGD demonstrates improved consistency, higher decision-support quality, and robustness to outlier/conflicting sources, highlighting its practical value for credible multi-source fusion.

Abstract

It is explored that available credible evidence fusion schemes suffer from the potential inconsistency because credibility calculation and Dempster's combination rule-based fusion are sequentially performed in an open-loop style. This paper constructs evidence credibility from the perspective of the degree of support for events within the framework of discrimination (FOD) and proposes an iterative credible evidence fusion (ICEF) to overcome the inconsistency in view of close-loop control. On one hand, the ICEF introduces the fusion result into credibility assessment to establish the correlation between credibility and the fusion result. On the other hand, arithmetic-geometric divergence is promoted based on the exponential normalization of plausibility and belief functions to measure evidence conflict, called plausibility-belief arithmetic-geometric divergence (PBAGD), which is superior in capturing the correlation and difference of FOD subsets, identifying abnormal sources, and reducing their fusion weights. The ICEF is compared with traditional methods by combining different evidence difference measure forms via numerical examples to verify its performance. Simulations on numerical examples and benchmark datasets reflect the adaptability of PBAGD to the proposed fusion strategy.

Paper Structure

This paper contains 12 sections, 1 theorem, 23 equations, 10 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Consistent credible fusion
  4. The conditional credibility
  5. Conditional credibility calculation
  6. The implementation of ICEF
  7. The PBAGD
  8. Experiment and application
  9. Initial event probability impact analysis
  10. Comparison on digital simulation
  11. Application
  12. Conclusion

Key Result

Theorem 1

The PBAGD satisfies the following properties:

Figures (10)

  • Figure 1: The flowchart of ICEF.
  • Figure 2: Results for the PBAGD.
  • Figure 3: Variation of $\alpha$ with varying $A_t$.
  • Figure 4: Variation of PBAGD with varying $A_t$.
  • Figure 5: Variation of PBAGD with varying $\alpha$.
  • ...and 5 more figures

Theorems & Definitions (8)

  • Example 1
  • Theorem 1
  • Proof 1
  • Example 2
  • Example 3
  • Example 4
  • Example 5
  • Example 6