Bayesian Update with Information Quality under the Framework of Evidence Theory

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Theory

Background:

• Classical Bayesian update is a standard for data fusion but neglects information quality.
• Integrating reliability assessments into Bayesian frameworks remains a challenge.
• Evidence theory offers a robust framework for handling uncertainty and information quality.

Purpose of the Study:

• To propose a generalized Bayesian update method that explicitly accounts for information quality.
• To enhance data fusion accuracy by incorporating source reliability.
• To extend the applicability of Bayesian updates in scenarios with varying information quality.

Main Methods:

• A novel Bayesian update framework is developed within evidence theory.
• Information quality is used to determine a discounting coefficient for prior probabilities.
• Discounted prior probabilities are transformed into basic probability assignments.
• Dempster's combination rule is applied to fuse information from multiple sources.
• Pignistic probability transformation converts the combined result to a posterior probability distribution.

Main Results:

• The proposed method effectively integrates information quality into the Bayesian update process.
• Demonstrated efficiency through a numerical example and a real-world target recognition application.
• The generalized method reduces to the classical Bayesian update when information quality is disregarded.

Conclusions:

• The novel Bayesian update method offers a significant advancement by incorporating information quality.
• This approach provides a more robust and accurate data fusion mechanism.
• The method serves as a generalized Bayesian update, applicable in diverse data fusion scenarios.