基于支持向量数据描述的预警技术及其应用

SVDD Based Early Warning Technique and Its Application

分析当前的主要预警方法,指出由于缺少非正常数据样本,使得现有的大部分预警方法不适用。为解决该问题,提出了基于核方法的支持向量数据描述预警技术。建立了一个用于检测非正常数据对象的一类分类器,检测数据对象是否在正常值超球体范围内。如果在超球体外,预警专家将最终确认这个数据对象是否为非正常的预警警兆。以广东省江门市的宏观区域经济数据为例,证明了该预警技术的有效性。

After reviewing the current early warning researches, this paper presents that most of current early warning methods are unsuitable because of lacking a historical ＂ill-represented＂ dataset. And then the support vector data description early warning technique based on kernel method is proposed to solve the problem. A one-class classifier is fitted to detect the ＂ill-represented＂ data objects by enclosing all ＂good＂ data objects in a hypersphere. If an object is outside the boundary of the hypersphere, an early warning expert would be prompted to decide whether the object is enough ＂ill-represented＂ for issuing a warning. An early warning experiment based on the macro-economic dataset of Jiangmen, Guangdong is conducted to verify the proposed technique.