基于代价敏感决策树的客户价值细分(英文)

A Cost-Sensitive Decision Tree Learning Model——An Application to Customer-Value Based Segmentation

由于错误分类代价差异和不同价值客户数量的不平衡分布,基于总体准确率的数据挖掘方法不能体现由于客户价值不同对分类效果带来的影响。为了解决错误分类不平衡的数据分类问题,利用代价敏感学习技术扩展现有决策树模型,将这一方法应用在客户价值细分,建立基于客户价值的错分代价矩阵,以分类代价最小化作为决策树分支的标准,建立分类的期望损失函数作为分类效果的评价标准,采用中国某银行的信用卡客户数据进行实验。实验结果表明,与传统决策树方法相比,代价敏感决策树对客户价值细分问题有更好的分类效果,可以更精确地控制代价敏感性和不同种分类错误的分布,降低总体的错误分类代价,使模型能更准确反映分类的代价。

The objective of this research is to extend the current decision tree learning model, to handle data sets with unequal misclassification costs. The research explores the issue of asymmetric misclassification costs through an application to customer-value based segmentation using empirical data collected from one of the largest credit card issuing banks in China. The data includes attributes from customer satisfaction survey and credit card transaction history is used to validate the proposed model. The results show that the proposed cost-sensitive decision tree for customer-value based segmentation is an effective method compared to the original decision tree learning model.