中文摘要：

提出一种基于数据集分割的极限学习机集成算法——DS—E—ELM．该算法主要包含以下3个步骤：首先，将数据集分成互不相关的k个子集，选择k-1个子集组合成一个训练集，这样可以得到k个不同的数据集；然后将新得到的k个数据集利用极限学习机训练得到k个分类器；最后对砖个分类器预测得到的结果通过多数投票的方法决定预测结果．通过对6个肿瘤数据集的实验证明，DS-E—ELM与单独的ELM、Bagging、Boosting等算法相比，具有更高的分类精度，且稳定性更好．

英文摘要：

In this paper, an Extreme learning machine ensemble method called DS-ELME, which is based on dataset splitting is presented. The DS-ELM-E method contains the following 3 steps, First, the training is divided dataset into k subsets, then k - 1 subsets are combined as a new training dataset, so we can get k different training dataset. Second, extreme leaning machine is used＇ to train the k different training dataset and to obtain k different classifiers. Third, the class label of the unknown data is predicted with the ensemble classifier through majority vote method. Experiments on six tumor datasets confirms that DS- E-ELM can obtain higher prediction accuracy compared with ELM, Bagging and Boosting, and more stable.