Multi-output extended belief rule-base system and its parameter learning schemes

As an advanced rule-based system, the extended belief rule-base (EBRB) system with single output has been applied in various areas due to its flexibility in knowledge representation. However, multi-output problems have not been adequately addressed in existing EBRB systems, although these problems are not unusual. In this study, an innovative multi-output EBRB (MO-EBRB) system is proposed with a unique inference scheme to handle multi-output problems. Also, a parameter learning scheme is designed to determine optimal parameter values in MO-EBRB system by constructing a multi-objective optimization model. The TOPSIS technique is used to select the optimal solution from a set of Pareto-optimal solutions generated by a nondominated sorting genetic algorithm. The effectiveness of the proposed system is demonstrated through its application in the auxiliary diagnosis of thyroid nodules. Comparison experiments indicate that the proposed MO-EBRB system could provide more accurate inference findings for the diagnosis of thyroid nodules compared to single output EBRB systems and other multi-output methods.