基于三元矩阵的层次分析多分类模型

A multi-classification model based on ternary code matrix and hierarchy analysis

针对支持向量机解决多分类问题时二分类向多分类扩展过程中的效率降低和数据集倾斜问题,提出了一种基于三元矩阵和层次分析的多分类模型的构造方法,优化支持向量机的多分类效果和效率,弥补1vs1、ECOC等主流算法的不足。该模型通过建立一种简单有效的获取样本集线性可分性构造分类器,从而减少支持向量机在处理多分类时的运算复杂程度。采用了UCI标准数据库中的Iris,Breast Tissue和Statlog等数据集对模型进行训练测试,测试结果表明所提出模型是有效的,尤其在大量数据下多分类的有效性。

Regarding to the problem of imbalance data set and poor classification efficiency that existed in current multiclass classification,a novel multi-class classification modeling method is proposed in this paper which inspired by ternary code matrix and hierarchy analysis method.The new modeling method gets the linear separability of the sample data,provides the basis for the classification model construction ,and then designs the multiclass model based on SVM.The novel multi-class classification modeling method can re-duce the number of binary classifiers and optimize support vector machine classification effectiveness and efficiency.At last we use datasets Iris ,Breast Tissue,and Statlog from UCI as examples to verify the cor-rectness of the proposed multi-classification model.Experiments show the correctness and efficiency of the novel multi-class classification method,especially show an excellent performance in big scale data multi-class classification problem.