摘要
为改进多分类器系统的性能,提出一个多分类器融合模型,该模型将和规则与多数投票作为特例纳入其体系中.用粒子群算法对融合模型进行优化得到PSO优化模型.在UC I标准数据集上对模型进行了实验研究.实验结果显示,同多数投票等6种融合方法中的最好结果相比,PSO优化模型使4个数据集上的错误率分别降低了91.44%、53.19%、5.76%、2.03%.实验中还发现,将性能较差的分类器从分类器集合中剔除能够进一步提高分类性能.
Aiming at improving the classification performance, a combination model of multiple classifier systems is presented, which takes the Sum rule and majority voting as its special cases. Particle Swarm Optimization（PSO）is used to select parameters for the model. An experimental investigation is performed on the UCI datasets and the encouraging results are obtained. The PSO combination model proposed in this paper is better than the other fusion methods given in the larger datasets with 91. 44 %, 53.19%, 5.76% and 2.03% reduction in error rate on 4 datasets, respectively. It is also shown that the rejection of weak classifier in the ensemble can improve classification performance further.
出处
《小型微型计算机系统》
CSCD
北大核心
2006年第7期1313-1316,共4页
Journal of Chinese Computer Systems
基金
国家"八六三"高技术研究发展计划基金项目(2003AA412020)资助.
关键词
模式分类
多分类器系统
融合模型
pattern classification
multiple classifier systems
combination model
PSO algorithm
