基于核方法的数据描述及其在企业关系评价中的应用

Study on Data Description based on Kernel Method and Its Application

数据描述又称为一类分类方法,用于描述现有数据的分布特征,以研究待测试数据是否与该分布相吻合.首先简要叙述了基于核方法的数据描述原理,指出:选择适当的核函数以及与之对应的参数,数据描述可应用于模式聚类中,并且这种聚类方法具有边界紧致、易剔除噪声的优势.针对基于数据描述的聚类方法在确定类别数目和具体样本类别归属上所存在的问题,提出了基于搜索的解决方法,理论分析和实例计算都验证了该方法的可行性.最后将该聚类算法应用到企业关系评价中,取得了较为合理的结果.

Data description, or one-class classification, is used to distinguish between a set of target objects and all other possible objects. On the basis of introducing the principle of data description based on kernel briefly, the paper points out that the data description can be used in pattern clustering if the kernel function and its parameters are choosed properly. The new clustering method has the advantage of tight boundary, so the disturbance of noise data is easy to elominate. Aiming at the difficulty of determing the number of clustering and which cluster each sample belong to in the new clustering method, the paper proposed a solution through searching, which has been proved theoretically. In the end of paper, the new clustering method is used in relation evaluation of enterprise. The result show that the new method is reasonable.