摘要
论文提出了一种新的多分类器系统模型,该模型通过定义两个基本分类器实时性能指标———实时决策支持度和实时决策置信度,在多分类器系统基础上构建临时的动态子系统,然后由该子系统代替初始多分类器系统来完成融合决策。动态子多分类器系统模型是一种不同于传统动态分类器选择和分类器联合方法的新模型,其能够更有效地排除不稳定基本分类器对多分类器系统融合决策性能的影响。试验表明该模型在模式识别性能上能够获得较好的性能,鲁棒性和可靠性比基本分类器和传统多分类器系统方法更强。
This paper proposes a new multiple classifier system model. Two real time performance indexes（real-time decision support and real-time decision confidence） of base classifier are defined in this new model. And a temporal sub system will be composed by these two real-time performance indexes for completing fusion decision instead of multiple classifier system. This dynamic sub system of multiple classi- fier system is a different method from the traditional dynamic classifier selection and classifier ensemble methods. And it can effectively ex clude the impact of unstable base classifier in fusion decision. Experiment results show that this new model can get higher accuracy than base classifier and some common used classifier ensemble methods.
出处
《计算机与数字工程》
2013年第6期911-913,983,共4页
Computer & Digital Engineering
基金
四川省科技计划项目(编号:2012ZZ0011)
广西师范学院科研启动经费
科学计算与智能信息处理广西高校重点实验室开放课题(编号:GXSCIIP201206)资助
关键词
分类器联合
多分类问题
模式识别
动态权重
实时决策支持度
实时决策置信度
classifier ensemble
multi-class classification problem
pattern recognition
dynamic weight
real-time decision support
real-time decision confidence
