摘要
为了平衡和减少两类不平衡数据的错分率,针对两类不平衡数据的分离超平面的偏置提出一种调整方法。该方法以两类错分概率相等为准则,使用特征提取方法,把高维样本投影到标准支持向量机的法向量上得到一维数据,当一维投影数据服从正态分布时,可由它所提供的信息,对标准的支持向量机中分离超平面的偏置进行调整。随机模拟试验表明了所调整的超平面不仅平衡了错分率而且减少了错分率。与现有方法相比,该方法具有较高的精度。
An adjustment method of the offset of separating hyperplane for binary-classification imbalanced data is proposed in order to balance and decrease the separating error ratio of the two classes of imbalanced data. The method is based on the rule of equal error probability. One-dimensional data set is obtained by using feature extraction that highly dimensional samples are projected on normal vector of the standard SVM. The adjustment of the offset of the separating hyperplane in standard SVM can be performed based on the information provided by the one-dimensional projective data set with normal distribution. The result of the random simulation experiment shows that separating error ratio is balanced and decreased. Compared with the existing methods, the method is of higher accuracy.
出处
《洛阳师范学院学报》
2008年第2期42-44,共3页
Journal of Luoyang Normal University
关键词
不平衡数据
正态分布
投影
支持向量机
偏置
imbalanced data
projection
normal distribution
support vector machines （SVM）
offset
