基于密度裁剪的SVM分类算法
|
摘要
针对SVM中当训练样本数非常多时,分类算法训练阶段的计算量庞大;当训练样本集中存在噪声时,影响训练阶段超平面划分的准确度,导致分类准确率降低等问题,提出了一种基于密度的训练样本裁剪的SVM算法。首先应用密度裁剪算法裁剪掉原始训练样本集中存在的噪声和冗余样本,作为新训练样本集,使用网格搜索算法在新的训练样本集中对SVM参数(C,g)寻优,进行SVM训练并建模,达到分类的目的。实验结果表明,该算法同无样本裁剪的SVM算法相比分类准确率有所提高、训练阶段的时间耗费大大降低。
Traditional SVM method has some problems in the practical application,its large computational demands and low classification accuracy when use it to deal with the large training samples set with noise.An improved SVM method is proposed for reducing the amount of training samples and the noise in data based on density.Firstly,reduce the amount of training samples and the noise in the data based on density,set the rest samples as the new training sample set,then use the grid searching algorithm to set optimal(C,g)parameters of the SVM,finally,create the SVM model for classification.The experimental results show that the algorithm is better than the SVM algorithm without cutting samples regardless of the classification accuracy or training time.
Traditional SVM method has some problems in the practical application,its large computational demands and low classification accuracy when use it to deal with the large training samples set with noise.An improved SVM method is proposed for reducing the amount of training samples and the noise in data based on density.Firstly,reduce the amount of training samples and the noise in the data based on density,set the rest samples as the new training sample set,then use the grid searching algorithm to set optimal(C,g)parameters of the SVM,finally,create the SVM model for classification.The experimental results show that the algorithm is better than the SVM algorithm without cutting samples regardless of the classification accuracy or training time.
引文
[1]Vapnik V N.统计学系理论的本质[M].张学工,译.北京:清华大学出版社,2000:5-22.
[2]陈永义.支持向量机方法应用教程[M].北京:气象出版社,2010:8-100.
[3]冯一宁,邵元海,陈静,等.基于层次聚类的大样本加权支持向量机[J].计算机工程与设计,2009,30(1):175-178.
[4]田新梅,吴秀清,刘莉.大样本情况下的一种新的SVM迭代算法[J].计算机工程,2007,(8):205-207.
[5]蔡艳艳,宋晓东.针对非平衡数据分类的新型模糊SVM模型[J].西安电子科技大学学报,2015,(5):120-124.
[6]李荣陆,胡运发.基于密度的KNN文本分类器训练样本裁剪方法[J].计算机研究与发展,2004,41(4):539-546.
[7]王艳飞,郝卫杰,范支菊,等.基于聚类和密度裁剪的改进KNN算法[J].青岛大学学报:自然科学版,2017,30(2):62-68.
[8]陈中杰,蒋刚,蔡勇.基于SVM一对一分类算法的二次细分法研究[J].传感器与微系统,2013(4):44-47.
[9]罗小燕,陈慧明,卢小江,等.基于网格搜索与交叉验证的SVM磨机负荷预测[J].中国测试,2017,43(1):132-135.
[10]徐晨,曹辉,赵晓.基于SVM的说话人识别参数选择方法[J].人工智能及识别技术,2012,(21):175-177.
[2]陈永义.支持向量机方法应用教程[M].北京:气象出版社,2010:8-100.
[3]冯一宁,邵元海,陈静,等.基于层次聚类的大样本加权支持向量机[J].计算机工程与设计,2009,30(1):175-178.
[4]田新梅,吴秀清,刘莉.大样本情况下的一种新的SVM迭代算法[J].计算机工程,2007,(8):205-207.
[5]蔡艳艳,宋晓东.针对非平衡数据分类的新型模糊SVM模型[J].西安电子科技大学学报,2015,(5):120-124.
[6]李荣陆,胡运发.基于密度的KNN文本分类器训练样本裁剪方法[J].计算机研究与发展,2004,41(4):539-546.
[7]王艳飞,郝卫杰,范支菊,等.基于聚类和密度裁剪的改进KNN算法[J].青岛大学学报:自然科学版,2017,30(2):62-68.
[8]陈中杰,蒋刚,蔡勇.基于SVM一对一分类算法的二次细分法研究[J].传感器与微系统,2013(4):44-47.
[9]罗小燕,陈慧明,卢小江,等.基于网格搜索与交叉验证的SVM磨机负荷预测[J].中国测试,2017,43(1):132-135.
[10]徐晨,曹辉,赵晓.基于SVM的说话人识别参数选择方法[J].人工智能及识别技术,2012,(21):175-177.