Kohonen's Map Approach for the Belief Mass Modeling

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Theory

Background:

• Traditional evidence theory struggles with mass attribution for compound hypotheses.
• Existing methods often fail to capture the full conceptual contribution of the theory.

Purpose of the Study:

• To propose an original method for estimating mass functions using Kohonen maps.
• To enable smart mass belief assignment for simple and complex hypotheses, modeling ignorance, imprecision, and paradox.

Main Methods:

• Utilizing Kohonen's map derived from the initial feature space and an initial classifier.
• Developing a method for basic belief assignment (BBA) suitable for large multivariate datasets.

Main Results:

• The proposed method achieves comparable or superior results to state-of-the-art BBA techniques on benchmark datasets.
• Demonstrated effectiveness in remote sensing image classification tasks.
• Successfully handles large volumes of data.

Conclusions:

• The novel Kohonen map-based approach enhances belief function estimation in evidence theory.
• This method offers a practical solution for mass function estimation with abundant multivariate data.
• The approach effectively models complex evidential scenarios, including ignorance and paradox.