基于权重阈值寻优的案例推理分类器特征约简

为提高案例推(case-based reasoning,CBR)分类器的分类准确率并降低时间复杂度,本文提出了一种基于权重阈值寻优的特征约简策略.首先通过基于数据驱动的方法对特征权重进行分配,得到每个特征的权重结果;其次,设计特征权重重要度阈值的适应度函数,并利用遗传算法对该重要度阈值进行优化搜索,最后根据得到的优化阈值与特征的权重分配情况,删除权重小于该阈值的特征从而完成特征的约简过程.通过对比实验,本文所提策略能够有效提高CBR分类器的分类准确率并降低时间复杂度,表明了权重阈值寻优约简策略的可行性与优越性.验证了本文方法不仅可以降低CBR分类器的时间复杂度,而且能够提高CBR的决策与学习能力.

Feature reduction method based on threshold optimization for case-based reasoning classifier

To improve the performance of case-based reasoning (CBR) classifier, we propose a feature reduction method based on threshold optimization for CBR classifier. First a data-driven method is adopted to conduct the feature weight distribution. Then, a weight threshold is introduced, where a genetic algorithm is utilized to obtain an appropriate threshold result, together with the feature weight and the threshold, the features of which the weights are lower than the threshold are deleted to accomplish the feature reduction process. The experimental results indicate that the weight distribution method and the threshold optimization method can improve the performance of CBR classifier, which confirms that the proposed reduction method is able to achieve a higher classification accuracy, decrease the time complexity, and improve the learning ability of CBR classifier.