摘要
针对多类分类问题提出了一种新的度量层分类器融合方法,为每个模式类设置多个决策模板,每个决策模板针对一种容易发生的分类错误,从而能够有效地降低错误率;此外,采用模糊系统表示Meta层样本与各个决策模板之间的关系,能够比较准确地计算样本属于各个模式类的总分类置信度。从公用数据仓库中选取了三个较大规模数据集对新方法进行测试,并且与k-近邻规则、投票法、朴素贝叶斯法、线性规则、模板匹配法等常用的分类器融合方法进行了比较。大量实验结果表明,对于类别数在3～15之间的分类问题,该方法具有较好的综合性能。
A new measurement level classifier combination method is proposed for multi-class problems.This method can decrease the classification error rate by representing each class with multiple decision templates corresponding to all kinds of classification errors.Moreover,in order to accurately calculating the overall confidences related to each class,fuzzy system is employed to describe the relationship between the meta-level samples and the decision templates.This method is tested on three large data sets selected from the public database and is compared with popular classifier combination methods such as voting,Naive Bayesian,linear combination rule,and decision templates.The experimental results show that the method performs well for the classification problems which include 3～15 classes.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第27期216-220,共5页
Computer Engineering and Applications
基金
国家自然科学基金(No.60773062
No.60903089
No.60801053)
河北省自然科学基金(No.F2009000215)
河北省教育厅科研计划项目(No.2008312)
北京市优秀博士学位论文指导教师科技项目(No.YB20081000401)资助项目~~
关键词
模式识别
分类器融合
模糊系统
pattern recognition
classifier combination
fuzzy system
