OCSVM集成模型解决支持向量机的大规模训练问题

Applying OCSVM Ensemble Model for The Problem of Training SVM with Large Scale Dataset

该文针对集成方法实现支持向量机大规模训练的相关问题进行了深入研究,提出了一种称为"DD-Boosting"的成员分类器产生算法,能够在大规模数据集情况下利用类似Boosting技术产生稳定、高泛化性能的成员分类器。在此基础上,推导出基于OCSVM的分类器集成模型,实验仿真表明,该集成模型能够获得比主投票方法更好的泛化性能,且通过调整正则参数避免了训练过拟合问题。

This paper firstly makes a deep study on the training problems on large-scale dataset by ensemble models.A novel algorithm called"DD-Boosting",which generates a pool of sub-classifiers like "Boosting",is proposed for large-scale dataset.Experiments shows that our algorithm can generate stable and accurate sub-classifiers,which are suitable for integrating.Meanwhile,a new ensemble model is deduced for integrating these sub-classifiers,which called "OCSVM".Preliminary experiments indicate that the new model's superiority on generalized performance than "Major Voting" method,which is partly because of the impacts of regularization parameter on avoiding over-training.