Belief rule base model considering dependencies between attributes

Abstract

Although the Belief Rule Base (BRB) established based on single attribute rules solves the problem of rule explosion, the dependency between attributes makes it impossible to accurately describe the relationship between attributes and results, resulting in low accuracy. For this problem, existing solutions are mostly limited by expert knowledge or only applicable to weakly correlated attributes. Therefore, a method of attribute clustering and intersection guided by particle swarm optimization is proposed, which integrates attributes linearly or non-linearly to obtain new attributes, which are used in the BRB model. This method can effectively solve the influence of attribute interdependence on BRB prediction results. Experiments show that the method has good accuracy in solving typical fault diagnosis, pattern recognition and other classification problems, and the number of rules has a linear relationship with the number of attributes and reference values, avoiding rule explosion.