基于可信度阈值优化的案例推理评价分类方法  

Trustworthiness evaluation method with threshold optimization for casebased reasoning classification

作　　者：严爱军[1,2,3] 赵辉[1,3] 王普[1,3,4] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]北京工业大学计算智能与智能系统北京市重点实验室,北京100124 [3]北京工业大学数字社区教育部工程研究中心,北京100124 [4]北京工业大学城市轨道交通北京实验室,北京100124

出　　处：《控制与决策》2016年第7期1253-1257,共5页Control and Decision

基　　金：国家自然科学基金项目(61374143);北京市自然科学基金项目(4152010)

摘　　要：为了提高案例推理(CBR)分类器的性能,提出一种基于可信度阈值优化的CBR评价分类方法.首先,通过一种可降低时间复杂度的改进型可信度评价策略对案例重用得到的建议解的可信度进行计算;然后,通过遗传算法(GA)对可信度阈值进行迭代寻优;接着,根据得到的优化阈值将目标案例及其建议解划分为可信集或不可信集;最后,对不可信集按多数重用原则进行分类结论的调整,从而实现可信的CBR评价分类.对比实验表明,改进的可信度评价策略能有效提高分类性能,从而可提高CBR分类器的决策与学习能力.To improve the performance of a case-based reasoning （CBR） classifier, a trustworthiness evaluation method with threshold optimization for case-based reasoning classification is proposed. Firstly, an improved trustworthiness evaluation（TE） strategy is adopted to calculate the trustworthiness value of the suggested solutions achieved in reuse step. Then, the optimal threshold va/ue of the trustworthiness is obtained by using the genetic algorithm（GA）. Subsequently, the target case and its suggested solution is divided into the trustworthy set and the untrustworthy set in accordance with this threshold value. Finally, the majority reuse strategy is adopted to adjust the suggested solutions in the untrustworthy set so as to fulfill an overall CBR evaluation classification process. The experimental results show that the proposed method can effectively increase the classification performance and improve the learning ability for a CBR classifier.

关 键 词：案例推理 评价分类 可信度阈值 遗传算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]