在客户流失预测中,引入损失函数、复杂性函数、属性维度函数和满意评价函数构建了两类错误满意控制模型结构,并以国内、国外电信公司客户流失预测为实例,发现该模型结构能在考虑预测分类错误成本、模型计算成本、属性规模的条件下,兼顾预测模型解释性,是研究客户流失两类错误问题的有效方法。
A tradeoff satisfatrory model structure of two types of errors was constructed by using loss function,complexity function,attribute dimensionality function and satisfaction estimation function for customer churn prediction.The structure was used in customer churn prediction for home and foreign telecommunication carriers.It is found that the structure includes classification error cost,model computation cost,attribute scale cost and explanation of prediction model,and provides an effective approach for studying on two types of errors for customer churn.