To improve the accuracy of customer churn prediction and ensure the quality of e-commerce platform services, a multi-feature fusion-based e-commerce platform customer churn prediction method is proposed.Firstly, the SMOTE algorithm is used to reduce data imbalance and construct a data set for predicting customer churn on e-commerce platforms; Then, select user characteristic attributes, analyse customer consumption behaviour based on consumption time characteristics, consumption value characteristics and consumption quantity characteristics, and integrate multiple characteristics of consumption behaviour; Finally, use the blending model and four base learners RF, GBTD, XGBoost and LightGBM, along with the secondary learner of the logistic regression model, to achieve customer churn prediction on e-commerce platforms.Experimental results have shown that the maximum error in predicting customer churn rate using the method proposed in this article does not exceed 0.1%, with an AUC value of 0.913 and a maximum accuracy of 0.92.
Ming Yang (Thu,) studied this question.