摘要
现有基于置信规则库的分类系统的分类准确率和效率受到系统参数设置以及规则库结构合理性的影响。为了寻找到最佳的参数值和最优的规则库结构,本文结合多目标免疫系统算法(multiobjective immune systemalgorithm,MISA)提出利用MISA多目标优化的置信规则库分类算法。该方法融合特征属性约简思想和差分进化算法思想建立训练模型,采用多目标免疫系统算法对系统复杂度和分类准确率进行多目标优化,从而寻找到分类模型的最优解。在实验分析中,首先将本文提出的置信规则库多目标分类系统MISA-BRM和置信规则库分类系统的实验结果进行对比,从复杂度和准确率两个维度说明本文方法的有效性。同时还将本文方法与现有的其他分类方法进行比较,验证本文方法的可行性和有效性。实验结果表明,本文方法能够有效地对基于置信规则库的分类系统的准确率和复杂度进行多目标优化。
The efficiency and accuracy of a belief rule base(BRB)classification system are limited to the determination of the systematic parameters and the structure of the rule base.In this study,we propose the usage of a BRB classification algorithm,i.e.,the multi-objective immune system algorithm(MISA),along with multi-objective optimization to determine the optimal parameters and the structure of the rule base.This method simplifies the characteristic attributes using a differential evolution algorithm to develop a training model and subsequently uses MISA to optimize the systematic complexity and the classification accuracy for identifying an optimal solution for the classification model.In the experiment,we initially compare the results of the BRM-based MISA(MISA–BRM)and those of the BRB classification system with respect to their complexity and accuracy to present the benefits of our method.Further,we compare the results with those of the existing classification methods to verify the feasibility and availability of the proposed method.The experimental results denote that the proposed method can effectively optimize the accuracy and complexity of the BRB classification system.
作者
林锦
胡家琛
刘莞玲
吴英杰
LIN Jin;HU Jiachen;LIU Wanling;WU Yingjie(College of Mathematics and Computer Science,Fuzhou University,Fuzhou 350116,China)
出处
《智能系统学报》
CSCD
北大核心
2019年第5期982-990,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(71501047,61773123)
福建省自然科学基金项目(2015J01248)
