摘要
提出了一种基于光滑正则的半监督支持向量机方法,并将其用于建立中小信用评级模型。它从少量标签样本和大量无标签样本中构造反映数据流形结构的光滑正则项,并结合到支持向量机的最大间隔分类器的学习过程。然后,提出一种渐进式方法来迭代获得"半标签"样本,稳健地提升支持向量机的泛化性能。在真实数据集上的实验结果表明,新方法获得的测试精度显著优于多种现有方法,非常适用于中小企业的信用评级任务。
This paper proposed a novel smooth regularization based semi-supervised support vector machine approach, and applied it to set up credit evaluation model of small-and-medium enterprises. It computes a manifold-related smooth regularization term on both few labeled samples and plenty of unlabeled samples, which is combined into the learning process of maximal margin classifiers. Then, it adopts a progressive method to acquire semi-labeled samples iteratively so that the generalization performance of support vector machine can be improved gradually. Experiments on reality dataset show that the testing accuracy of proposed approach outperforms several popular ones, and is very suitable for evaluating credit grades of small-and-medium enterorises.
出处
《计算机科学》
CSCD
北大核心
2013年第10期239-242,共4页
Computer Science
基金
教育部人文社会科学研究一般项目(10YJCZH153)资助
关键词
支持向量机
信用评级
半监督学习
光滑正则
Support vector machines, Credit evaluation, Semi-supervised learning, Smooth regularization
