Non Parametric Fuzzy Modeling of Belief Functions in Evidence Theory

M. Arif, T. Brouard, and N. Vincent (France)

Keywords

DempsterShafer evidence theory, modeling of belief functions, fuzzy membership function, distance classifiers.

Abstract

Dempster-Shafer evidence theory has been reported with good performances on combining information coming from several sources in different areas. Its performance depends considerably on the function which is employed as function of allocation of mass of belief. There exists some work in this area in the literature. In the present work, we propose a prototype based fuzzy modeling for belief functions in Dempster-Shafer (DS) evidence theory. Our modeling approach as compared to all other existing approaches has the main advantage that it is a non parametric and it can contribute towards a generic approach. We are employing DS evidence theory for a pattern recognition problem. We validate our methodology in a recognition process of an incoming pattern at the decisional step by combining information from two trained distance classifiers considering k nearest neighboring prototypes of the incoming pattern. For performance confirmation of our developed fuzzy approach, a comparison was realized with an existing proven efficient fusion method relying on the confusion matrix of classifiers and belief integration based on Bayesian formula. Results obtained are thought encouraging.

Important Links:



Go Back