基于混合样本训练的并行层叠支持向量机研究

Research on Parallel Casecade SVM Based on Mixed Samples

层叠支持向量机将原始数据集随机划分为多个子集,对数据子集采取并行训练,可以有效提高分类器的训练效率。但其在将原始数据随机划分为多个训练子集时,可能会给各并行节点带来文本信息结构的不均衡,进而影响分类器的最终分类效果。提出了一种基于混合样本训练子分类器的训练模型,实验表明,基于混合样本训练的层叠支持向量机,可以较好地解决训练样本信息结构不均衡问题,保证层叠训练得到的分类器具有较好的精确度和稳定性。

Cascading support vector machine(SVM)divides the original data sets by randomly dividing them into multiple subsets.Parallel data subset training can effectively improve the training efficiency of the classifier.However,when the original data is randomly divided into multiple training subsets,it may bring the imbalance of various parallel node text information structure to each parallel node,and then affect the classification effect of the final classifier.In this paper,a training model based on mixed sample training subclassifier is proposed,and the experiment shows that the cascade support vector machine based on mixed sample training can solve the problem of unbalanced information structure of training samples,and ensure that the classifier obtained by cascade training has better accuracy and stability.