一种代价敏感学习方法在电信业流失预测中的应用

根据已有的流失预测方法,提出新的流失预测方法解决数据挖掘中的非对称错分代价问题.该方法以传统C4.5决策树算法为基准分类器,融合代价调整方法实现代价敏感学习.相比之下,C4.5决策树算法仅是基于样本错分代价相同假定,建立了一种错分率最低而非总错分代价最低的预测模型.基于某电信企业的客户数据,及流失客户和非流失客户代价非对称的实际,实证研究结果表明,CS-C4.5通过调整流失类和非流失类样本的比例,大大降低了传统分类算法的样本错分总代价.该方法对于提高电信企业的核心竞争力具有重要的现实意义.

Application of a cost-sensitive method for churn prediction in telecommunication industry

To deal with the data mining problem of asymmetry misclassification cost,an innovative churn prediction method is proposed based on existing churn prediction research.This method adjusts the misclassification cost based on the C4.5 decision tree as a baseline classifier,which can obtain the prediction model with a minimum error rate based on the assumption that all misclassifications have the same cost,to realize cost-sensitive learning.Results from customer data of a certain Chinese telecommunication company and the fact that the churners and the non-churners have different misclassification costs demonstrate that by altering the sampling ratio of churners and non-churners,this cost-sensitive learning method can considerably reduce the total misclassification cost produced by traditional classification methods.This method can also play an important role in promoting core competence of Chinese telecommunication industry.