基于假设间隔的弱随机特征子空间生成算法

On Generating Algorithm of Weak Random Subspace Based on Simba

集成算法是机器学习领域的研究热点。随机子空间算法是集成算法的一个主要算法。随机子空间生成的特征子集可能含有冗余特征、甚至噪声特征,影响算法的分类精度。为此,提出了一种基于假设间隔的弱随机特征子空间生成算法(WRSSimba),有效去除了特征子集中冗余特征和噪声特征。在UCI数据集上的实验结果表明,WRSSimba的分类性能优于随机子空间算法和Simba算法。

The integration algorithm is a hot research field of machine learning, and random subspace algorithm is a major algorithm of integration algorithm. But feature subset generated by random subspace may contain redundant feature and even noisy feature that affects the accuracy of classification. Therefore, in this paper, a new algorithm of weak random subspace based on Simba（WRSSimba） is introduced. This algorithm effectively eliminates the redundancy and the noisy feature of feature subspace. The experimental results on UCI datasets show that, its performance of classification is better than random subspace algorithm.