摘要
针对传统的集成在线顺序极限学习机在集成决策时忽略各个学习机的分类性能而平均设置集成权重,导致集成系统分类准确率较低且不稳定,就此提出一种新的集成规则。在改进的方法中,首先根据每个在线顺序极限学习机在验证集上的分类准确率对被集成的学习机进行筛选;其次,再根据每个被集成进来的学习机在验证集上的分类准确率设置该学习机的集成投票权重。在四个标准分类数据集上的实验结果表明,本文提出的集成在线顺序极限学习机能够以更高的稳定性获得更高的分类准确率。
Since traditional ensemble of online sequential extreme learning machine ignores the classification performance of each extreme learning machine and selects the same ensemble weight in ensemble decision, which leads to worse classification accuracy and stability, a modified ensemble regulation is proposed in this paper. In the modified approach, firstly, the integrated learning machines are chosen according to the validate accuracy of each online sequential extreme learning machine. Then the voting weights are assigned in accordance with the validate accuracy of integrated online sequential extreme learning machines. Experiment results on four benchmark classification data sets verify that the proposed method could obtain better classification accuracy with the better stability than OS -ELM and EOSELM.
出处
《无线通信技术》
2013年第3期39-44,共6页
Wireless Communication Technology
基金
国家自然科学基金(Nos.61271385
60702056)
江苏省自然科学基金(No.BK2009197)项目
关键词
在线顺序极限学习机
集成
准确率
稳定性
online sequential extreme learning machine
ensemble
accuracy
stability
