非均衡数据的最小二乘支持向量机阈值新算法

摘要

Analyzing the problem that traditional Least Square Support Vector Machine (LSSVM) classifier is biased to the larger class when the data are unbalanced. An improved method of BLSSVM is proposed. Based on linear discrimination, the mean and variance of the projected class are obtained by projecting two classes of samples onto the normal vector of the classification hyperplane, then the threshold of the hyperplane is adjusted, according to the principle that error probability of two classes are equal. The proposed algorithm can compensate the ill-effect of tendency and improve the accuracy. Simulations on imbalaneed artificial data and real data show the feasibility and validity of the proposed method.%针对传统的最小二乘支持向量机对于非均衡数据的分类时,分类结果具有对较大类数据的偏向性问题,为了减小分类器的负担和样本的错误率.提出一种新的最小二乘支持向量机阈值计算方法进行修正.根据线性判别思想,计算出两类样本的在分类超平面法向量上的投影点的均值和方差,依据对两类样本错分概率相等准则,给出新的阈值计算方法从而实现对超平面进行调整.方法可补偿最小二乘支持向量机对非均衡数据分类的倾向性,提高其预测分类精度.通过数值仿真实验表明了方法的可行性与有效性.