一种基于自适应证据推理规则的集成学习方法

Ensemble learning method based on adaptive evidential reasoning rule

当前集成学习中的结合策略难以兼顾各个基学习器之间的信息和模型的可解释性。使用证据推理(evidential reasoning,ER)规则作为结合策略,将各个基学习器结果作为证据参与融合,可以较好地解决以上问题。但传统ER规则的证据参数是单一的,对不同的基学习器模型使用相同的证据参数显然是不合理的。为此,提出一种基于自适应证据推理(adaptive-evidential reasoning,A-ER)规则的集成学习方法,该方法在每次证据融合前对证据的类别进行判断,针对不同的证据类别自适应分配不同的证据参数。通过不同的分类案例表明,该方法与案例中其他方法相比具有更高的分类精度,证明了该方法使证据参数设置更加合理且具有更好的可解释性和泛化能力。

Current ensemble learning combination strategies have difficulty in accommodating information and explainability of models among base learners.Using evidential reasoning(ER)rules as a combination strategy,the above problems can be solved by involving the results of each base learner as evidence in the integration.However,the evidence parameters of conventional ER rule are single,and it is obviously unreasonable to use the same evidence parameters for different base learner models.This paper proposed an ensemble learning method based on adaptive evidential reasoning(A-ER)rule.This method judged the type of evidence before each evidence fusion and adaptively assigned the different evidence parameters according to different types of evidence.Results from different classification cases show that the proposed method has a higher classification accuracy compared to other methods in the cases.It proves that this method has more reasonable evidence parameter settings and superior interpretability and generalization ability.