论文部分内容阅读
针对支持向量机中两类不平衡数据的分离超平面提出一种调整算法.首先用标准的支持向量机对原始数据进行初步训练,产生一个分离超平面的法向量.然后把高维样本投影到该法向量上得到一维数据.最后由投影数据的标准差以及样本容量所提供的信息,给出两类数据惩罚因子比例,再用标准的支持向量机进行第2次训练,从而得到一个新的分离超平面.实验显示该方法的有效性,即在一般情况下能平衡错分率,甚至还能减少错分率.
Aiming at the separation hyperplane of two types of unbalanced data in SVM, an adjusting algorithm is proposed.Firstly, a standard SVM is used to train the original data to generate a normal vector that separates the hyperplane.Then the high dimensional samples are projected onto The normal vector to obtain one-dimensional data.Finally, the standard deviation of the projection data and the information provided by the sample size to give the proportion of two types of data penalty factor, and then use the standard support vector machine for the second training to get a new The experimental results show that this method is effective in balancing the misclassification rate and reducing the misclassification rate.