SVM多类别分类方法在客户流失预测中的应用研究  

ON APPLYING SVM MULTI-CLASS CLASSIFICATION IN CUSTOMER CHURN PREDICTION

作　　者：李兴国[1] 谢伟[1] 卢光松[1] 

机构地区：[1]合肥工业大学管理学院,安徽合肥230009

出　　处：《计算机应用与软件》2010年第3期94-96,113,共4页Computer Applications and Software

基　　金：安徽省教育厅项目(2006KJ043B)

摘　　要：针对移动客户的多类别特点以及不同类别客户的数据集分布不平衡,把CW-SVM应用于多类别分类,提出了DAG-CWS-VM算法,分别用1-V-R SVM、1-V-1 SVM、DAG-SVM与DAG-CWSVM算法进行分类预测,并用ROC曲线、AUC值、提升度(lift)指标对四种模型进行评价。实验表明DAG-CWSVM算法不仅能够处理数据多类别分类问题,而且能够有效地解决数据集不平衡对预测结果造成的影响,具有较好的预测效果。In light of multi-class character of China Mobile customers and the imbalanced datasets distribution of the customers in different classes, the class weighted SVM model CW-SVM was applied to muhi-class classification, a DAG-CWSVM algorithm was proposed. Classifica- tion was predicted by 1-V-R SVM, 1-V-1 SVM, DAG-SVM and DAG-CWSVM respectively, and the indices of ROC （ receiver operating characteristic curve） ,AUC（ area under ROC curve） value and lift were employed to assess the above four models. It was demonstrated by the test that DAG-CWSVM algorithm was suitable for dealing with multi-class classification of the data,and it can also effectually overcome the impact on prediction results caused by the imbalance of datasets with higher prediction effect.

关 键 词：支持向量机 多类别客户流失预测 AUC 提升度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]