可变惩罚系数比例的分类SVM模型

A Classification SVM Model with Variable Punishment Coefficient Ratio

为了解决常规分类SVM模型在样本比例相差较大情况下的不适用问题,提出了可变惩罚系数比例的分类SVM模型。在所提出的模型中,惩罚系数的比例以其对应样本数的反比来确定。最后,在90¨10样本比例下,分别用常规SVM模型和所提出的SVM模型得到相应的分割线,并将其进行比较,实验结果验证了本模型的有效性。虽然本研究所采用的核函数是最简单的向量内积,但所提出的模型对于所有的核函数都是适用的。

In order to solve the problem that normal classifier SVM model was unsuitable with the circumstances that samples ratio was very large,a classification SVM model with variable punishment coefficient ratio was proposed. In the new model,the ratio of punishment coefficient was inversely proportional to the ratio of corresponding sample sizes. At last ,experiments were conducted. Segmentation lines were respectively obtained through these models, and the results verified the new one. In this research, the kernel function is in the simplest form of inner product ,but the new model is applicable for all the kernel functions.