提高案例推理分类器的可靠性研究

摘要

To aim at the reliability issue of case-based reasoning (CBR) classifier, improved strategies for case retrieve and case reuse are introduced, respectively. In the retrieve step, a new attribute weight assignment method based on the water-filling principle is proposed to optimize the feature weight; particularly, the Lagrange function is constructed by utilizing the mean value and the standard deviation of each attribute to achieve the weight result, then a weight threshold is set to conduct the attribute reduction. In the reuse step, a confidence-reuse strategy is introduced to improve the efficiency of the classifier by calculating the confidence of the target case that belongs to each class. Simulation experiments show that the proposed methods could increase the classification accuracy and efficiency, which proves that the improved strategies could effectively enhance the reliability of the CBR classifier.%针对案例推理(Case-based reasoning, CBR)分类器的可靠性问题,本文提出一种改进的案例检索和案例重用方法。首先在案例检索环节应用注水原理对属性权重进行优化分配,利用每个属性数据的标准差和均值构造拉格朗日函数求得属性权重,并设定重要度阈值指导属性约简；其次在案例重用环节引入基于可信度的重用策略,通过计算目标案例分属于各个类别的可信度大小来确定当前案例的分类结果。最后通过实验对比,表明本文方法能有效提高分类精度和效率,分类器的可靠性得以保障。