支持向量分类机阈值的唯一化

The Unique of Threshold for Support Vector Classifiers

支持向量分类机是数据挖掘的新方法,它对应于一个凸二次规划,该规划的不唯一解带来阈值不唯一问题。针对阈值不唯一问题,研究了在具体应用过程中如何修改模型参数,在不影响具体应用问题解决的前提下,提出使阈值唯一化的一个解决方法,同时给出参数变化后最优解的理论结果。这种唯一化的方法不仅为支持向量分类机数据扰动分析新的研究方向作基础准备,而且可以克服由不唯一阈值构成的多决策函数在实际应用问题中带来的困扰。

Support Vector Classifiers （SVC） is a new method for data mining. It is equal to a quadratic optimal problem. The various solutions cause the non-uniqueness of threshold problem. A method which can make the threshold uniqueness without changing the primal applied model is introduced. In addition, this paper establishes theory for the threshold uniqueness. The theory is a foundation for SVC perturbation analysis field. In addition, the methods can be applied to solve the trouble of many decision functions.