A new basic probability assignment generation and combination method for conflict data fusion in the evidence theory.

Area of Science:

• Artificial Intelligence
• Information Fusion
• Decision Theory

Background:

• Dempster-Shafer evidence theory is widely used for information fusion.
• The Dempster's combination rule faces challenges with fusion paradoxes.
• Existing methods for generating basic probability assignments (BPAs) require improvement to handle these paradoxes.

Purpose of the Study:

• To propose a new method for generating BPAs to address fusion paradoxes in Dempster-Shafer theory.
• To improve the reliability and accuracy of information fusion.
• To enhance the performance of classification tasks using the improved fusion method.

Main Methods:

• A novel BPA generation method integrating cosine similarity and belief entropy.
• Utilizing Mahalanobis distance to measure similarity between test samples and focal elements.
• Adjusting BPAs based on reliability (cosine similarity) and uncertainty (belief entropy).
• Applying Dempster's combination rule for fusing the adjusted BPAs.

Main Results:

• The proposed method effectively resolves classical fusion paradoxes.
• Numerical examples demonstrate the method's effectiveness in overcoming paradoxes.
• Classification experiments show improved accuracy rates, verifying the method's rationality and efficiency.

Conclusions:

• The new BPA generation method successfully tackles fusion paradoxes in Dempster-Shafer theory.
• The approach enhances information fusion by providing more reliable and accurate BPAs.
• This method offers a robust solution for classification problems and other information fusion applications.