A data driven rule-base inference approach for classification systems

This paper proposes a generic data driven inference methodology for rule-based classification systems. The generic rule base is in a belief rule base structure, where the consequent of a rule takes the belief distribution form. Other knowledge representation parameters such as the weights of both input attributes and rules are also considered in this framework. In an established rule base, the matching degree of an input between the antecedents of a rule is firstly computed to get the activation weight for the rule. Then a weighted aggregation of the consequents of activated rules is used for the inference process. Two numerical examples are provided to illustrate the proposed method.