A New Approach to Generate Basic Probability Assignment in the Open World

Generalized evidence theory is an effective mathematical framework for addressing multi source information fusion problems. It is particularly adept at handling uncertainty, even in open-world scenarios. A crucial component of GET is the generalized basic probability assignment (BPA), which is used to quantify the belief assigned to different hypotheses. However, determining an accurate BPA remains a challenging and open problem, because it influences the result of evidence fusion. In this paper, we propose a novel method for generating BPA based on Gaussian fuzzy number (GFN). First, we construct a multi-subset representation model using GFN. Next, BPA are generated using this model along with specific strategies to ensure accuracy. Finally, the effectiveness is proved through experiments on the Iris dataset.