基于改进的K近邻和支持向量机客户流失预测

Telecom customer churn prediction based on improved K-nearest neighbor and support vector machine

将K近邻分类法和支持向量机分类法结合起来,给出一种电信客户流失预测方法,即对边界样本采用加权K近邻分类,而对非边界样本采用改进的支持向量机分类。在公开不平衡数据集和电信数据集上的实验可验证所给方法有效,且能提高少数类的检测精度和总体评价指标。

Combining K-nearest neighbor classification and support vector machine classification,a telecom customer churn prediction method is presented.According to this method,the boundary samples are classified by the weighted K-nearest neighbor,and the non boundary samples are classified by the improved support vector machine respectively.Experiments on open unbalanced data sets and telecommunication data sets show that,the proposed method is effective and can improve the overall evaluation index,especially improve the detection accuracy of the minority.