基于Bagging的概率神经网络集成分类算法  

Bagging-based Probabilistic Neural Network Ensemble Classification Algorithm

作　　者：蒋芸[1] 陈娜[1] 明利特[1] 周泽寻[1] 谢国城[1] 陈珊[1] 

机构地区：[1]西北师范大学计算机科学与工程学院,兰州730070

出　　处：《计算机科学》2013年第5期242-246,共5页Computer Science

基　　金：国家自然科学基金项目(61163036;61163039);甘肃省科技计划(甘肃省自然科学基金项目1010RJZA022;1107RJZA112);2012年度甘肃省高校基本科研业务费专项资金项目;甘肃省高校研究生导师项目(1201-16);西北师范大学第三期知识与创新工程科研骨干项目(nwnu-kjcxgc-03-67)资助

摘　　要：目前的神经网络较多集中在以BP算法为基础的BP神经网络上。针对BP神经网络的不足,在分析研究概率神经网络和机器学习的基础上,结合集成学习的思想,提出了基于Bagging的概率神经网络集成分类算法。理论分析和实验结果都表明,提出的算法能够有效地降低分类误差,提高分类准确率,具有较好的泛化能力以及较快的执行速度,能够取得比传统的BP神经网络分类方法更好和更稳定的分类结果。Neural networks classification algorithm now more concentrates on the BP algorithm which is the representa- tive of the neural networks. Considering the disadvantage of BP neural network, based on the analysis of probabilistic neural networks and machine learning, and combining with the idea of ensemble learning, we proposed a new classifica- tion algorithm which is probabilistic neural networks ensemble based on Bagging. Theoretical analysis and experimental results show that the proposed algorithm can effectively reduce the classification error and improve accuracy of classifi- cation. The proposed algorithm has good generalization ability and faster speed of execution than the traditional classifi- cation methods such as BP neural networks and it can achieve better and more stable classification result.

关 键 词：分类 BP神经网络 概率神经网络 集成学习 BAGGING 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]