基于重采样与属性约简的多模态选择性集成学习

Multimodal selective ensemble learning based on resampling and attribute reduction

通过对重采样技术和属性约简方法进行研究,提出一种多模态选择性集成学习算法SE_RSAR。采用重采样方法扰乱样本空间,采用一种基于相对决策熵的属性约简方法扰乱特征空间,通过这种多模态的扰乱策略增加个体分类器之间的差异性。实验在多个UCI数据集上完成,KNN算法被用来训练个体分类器。实验结果表明,相对现有的集成学习算法,SE_RSAR算法能够取得更好的分类效果。

Through the research on resampling technology and attribute reduction method,a multimodal selective ensemble lear-ning algorithm SE_RSAR was presented.The resampling method was used for the perturbation of sample space and a RDE(i.e.,relative decision entropy)based attribute reduction method was used for the perturbation of feature space.This multimodal perturbation strategy was used to increase the difference between the individual classifiers.The experiments were performed on multiple UCI datasets,and the KNN algorithm was used to train the individual classifiers.Experimental results show that compared with the existing ensemble learning algorithms,the SE_RSAR algorithm can achieve better classification results.