摘要
针对决策模板法在业务感知准确率上的局限性问题,提出了加权决策模板法。该方法首先利用有监督的神经网络模糊聚类分类器作为基本分类器,再通过混淆矩阵衡量分类器对样本不同类别的置信度,经过两级的性能权衡,赋予该算法更高的可信度。在训练阶段根据错误分类的样本构造一个附加的加权决策模板,若在测试阶段有样本离该模板的距离最近时,可以认为该样本被错误分类的可能性很大,从而保证该算法具有高识别准确率。实验结果表明,与决策模板法对比,加权决策模板法在业务感知上具有更高的准确性。
Aiming at the limitation of recognition accuracy that the decision template method used on service awareness,it puts forward the weighted decision template method. This method firstly uses supervised fuzzy clustering neural networks classifier as the basic classifier, and then through the confusion matrix measures classifier samples for different categories of confidence, after two stage performance trade-offs, gives higher credibility of the algorithm. In the stage of training, it constructs an additional weighted decision template according to the error classification of the samples, if in the testing phase, there are the shortest distance from samples to template, it thinks the sample in a high probability of error classification,so that the algorithm has high recognition accuracy. Experimental results show that, compared with that decision template method, the weighted decision template method has higher accuracy on service awareness.
作者
杨应雷
周金和
王川潮
YANG Yinglei;ZHOU Jinhe;WANG Chuanchao(School of Information and Telecommunication Engineering, Beijing Information Science and Technology University, Beijing 100101, China)
出处
《计算机工程与应用》
CSCD
北大核心
2017年第2期118-123,共6页
Computer Engineering and Applications
基金
国家自然科学基金(No.61271198)
北京市自然科学基金(No.4131003)
北京市教委科技计划重点项目(No.KZ201511232036)
关键词
决策模板法
模糊聚类
业务感知
分类器
decision template method
fuzzy clustering
service awareness
classifier
