基于耦合模拟退火S3VM的信用预测

Coupled simulated annealing semi-supervised SVM for credit prediction

在标记数据不足的情况下,半监督支持向量机(S3VM)可以有效利用标记数据和未标记数据提高模型性能。针对传统模拟退火S3VM方法在低温时容易陷入局部最优的问题,提出将耦合模拟退火用于半监督支持向量机的超参数选取,即CSAS3VM方法,并应用到信用预测中。在爬取的企业信用和UCI的个人信用两种数据集上与7种已有的方法进行对比实验,精度和F-1值两项指标的实验结果表明,提出的CSAS3VM方法明显优于模拟退火半监督支持向量机和其它传统方法,并且在5组包含均衡和不均衡的数据集上表现稳定。

In the case of small amount of labelled data,semi-supervised support vector machine(S3VM)can effectively utilize the labeled data and unlabeled data to improve the performance of the model.Aiming at the problem that the traditional simulated annealing S3VM method is easy to fall into local optimum at low temperature,the coupled simulation annealing was proposed for the semi-supervised support vector machine parameter selection,namely CSAS3VM method,and applied to credit prediction.Compared with seven existing methods on both the crawled corporate credit and UCI’s personal credit datasets,the experimental results of two indicators of precision and F-1 scores show that the CSAS3VM proposed is obviously superior.In addition,CSAS3VM performs stably well under five subsets of experiment data including imbalance and balance datasets.