基于代价敏感SVM的电信客户流失预测研究

针对客户流失数据集的非平衡性问题和错分代价的差异性问题，将代价敏感学习应用于Veropoulos提出的采用不同惩罚系数的支持向量机，建立客户流失预测模型，对实际的电信客户流失数据进行验证。通过与传统SVM、C4.5和ANN对比研究，结果显示此方法在精确度、命中率、覆盖率和提升度均有所改善，表明此方法有效地解决了数据集的非平衡性和错分代价问题，是进行客户流失预测的有效方法。

Study of telecom customer churn prediction based on cost sensitive SVM

To deal with the problem of unbalanced data classification and asymmetry misclassification cost in customer churn prediction,applied cost sensitive learning to the improved SVM which Veropoulos suggested it could handle the problem unbalanced data classification well to the model of customer churn prediction. The cost sensitive SVM was compared with traditional SVM, C4.5 and ANN through real telecom customer churn data. And found that it has a distinct improvement in accuracy rate, hit rate, covering rate and lift coefficient. It can be used as an effective measure for customer churn prediction.