Abstract:

Considering on the difficulty of interpretable reasoning for group decision in the mislabeled classification context, a method of evidential chains (ECs)-based reasoning for group decision analysis is proposed. Based on the consistency and convexity of the belief function, association similarities between the query case and ECs on their attributes are used as the weights of the multi-source information, and the mixed integer optimization model of ECs-based fusion reasoning is formed to maximize the reasoning performance, achieving the most closely related ECs. For reasoning with the mislabeled instances, this framework facilitates belief preference to induce the conclusion of queries. A diagnostic experiment with multi-source sensory data verifies the efficiency and rationality of the method.