摘要
改进传统的基于二叉树结构的支持向量机多类分类方法。将无监督聚类引入到算法中,利用无监督聚类剔除大量的非支持向量样本,同时对于无监督聚类在异类样本相近时出现的性能下降问题,引入线性判别分析使得同类样本聚集,异类样本分散,确保聚类精度。线性判别分析和无监督聚类结合能够显著地缩减训练样本。该方法能够在保持分类准确率的情况下有效地提高SVM的分类速度。
To improve traditional multi-class SVM method based on binary-tree.Using unsupervised clustering to extract training set,meanwhile,using linear discriminant analysis to solve the performance degradation of clustering when samples in different classes are similar,makes the samples in the same classes are gathered together and the samples in different classes are scattered,to ensure the accuracy of clustering.LDA and cluster can reduce training sample efficiently.The approach improves the speed of classification effectively while maintaining classification accuracy.
出处
《西北大学学报（自然科学版）》
CAS
CSCD
北大核心
2014年第4期559-562,共4页
Journal of Northwest University（Natural Science Edition）
基金
国家自然科学基金资助项目(61070176)
关键词
支持向量机
线性判别分析
模糊C均值聚类
多类分类
二叉树
support vector machine
linear discriminant analysis
fuzzy C-means clustering
multi-class classification
binary-tree
