互联网金融企业用户流失预测特征提取方式对比研究

Comparing Feature Constructing Methods for Customer Churn Prediction: A Research in the Field of Internet Finance

近年,互联网金融企业快速发展后用户流失问题变得越来越重要,而精准的用户流失预测能为企业制定相关策略提供决策支持。本文通过一家互联网金融公司的用户数据,包含基本信息、交易信息和日志行为信息,使用RFM(recency-frequency-monetary)和TFPD(time-frequency plane domain)法提取用户行为特征和交易特征,并使用逻辑回归、随机森林和支持向量机建立流失预测模型。结果表明:用户行为和交易数据通过RFM方式提取特征在逻辑回归上有较好的表现;用户交易数据通过TFPD方式提取特征在随机森林和支持向量机模型上有较好的表现。最后,本文综合用户基本特征和通过最优提取方式提取的行为与交易特征训练用户流失预测模型,案例企业应用该模型可提前识别有流失风险的用户,并降低用户召回成本。

The emerging and popularity of Internet finance in recent years make customer churn prediction serious because it can provide company with decision support.In this paper,we use real data from an internet finance company,apply and compare three effective models to make customer churn prediction.We use recency-frequency-monetary(RFM)analysis and time-frequency plane domain(TFPD)analysis to process users trading and behavioral data for constructing features.Then,we use the constructed features in three classification models-logistic regression,random forest,and support vector machine to make comparisons of RFM and TFPD for predictive performance on the customer churn.The results show that RFM performs better in addressing all two types of data by logistic regression model,while TFPD gains better results in addressing trading data by random forest and support vector machine models.Finally,we combine the features constructed by correct schema with base features to train customer churn prediction models.In this research,the results is meaningful for the case company because it can help the company find churn customers and reduce its customer recall cost.