基于GMDH的“一步式”客户流失预测集成建模

“One-step” customer churn prediction ensemble modeling based on GMDH

在客户流失预测问题中,客户数据的特征往往会影响模型的预测效果.分析了常用的"两步式"客户流失预测方法的不足,提出了基于数据分组处理(GMDH)技术的"一步式"客户流失预测集成研究框架.该框架一方面将数据预处理和客户流失预测建模过程进行集成,另一方面用多分类器集成策略进行客户流失预测建模.以客户数据类别分布不平衡的客户流失预测问题为例,构建了与数据特征相适应的"一步式"集成模型.实证结果表明,该方法能够更有效地进行客户流失预测.

In customer churn prediction,the characteristics of customer data tend to affect the prediction results of modeling.After analyzing the disadvantages of the commonly used ＂two-step＂ methods,this paper proposed ＂one-step＂ ensemble framework for customer churn prediction based on group method of data handling（GMDH）.On the one hand,this framework fused data pre-processing and customer churn prediction modeling;on the other hand,it adopted multiple classifiers ensemble strategies to model the customer churn prediction.Regarded the customer churn prediction problem with imbalanced data as an example,a ＂one-step＂ ensemble model corresponding to the customer data characteristic was constructed. Empirical results shown that this method can predict customer churn more effectively.