Customer Churn Pre‐identification Method Based on Feature Engineering and Stacking Strategy

10.6911/WSRJ.202206_8(6).0088

Yuansheng Song；Deyi Gong；Liangao Guo；Wanran Bu

Data mining ； Feature analysis ； Ensemble learning ； Prediction model ； SMOTE algorithm

World Scientific Research Journal

8卷6期（2022 / 06 / 01）

677 - 684

英文

With the rapid development of economy, commercial activities become more and more frequent. The customer has become the position that the enterprise and commercial activity scramble for, and the direction of the customer also decides the direction of the business. In order to sustain profits, business companies and enterprises should capture and retain customers, identify the customers who will be lost in advance, and make timely adjustments to retain them reasonably. The existing prediction models for churn are relatively single or simple and inefficient in the face of high‐dimensional data. In this paper, a model fusion strategy based on customer churn pre‐identification method is proposed. XGBoost and LightGBM were used as the first layer models, and LogisticRegression was used as the second layer model to prevent overfitting, Stacking strategy was used to build the model in this paper. By comparing the commonly used XGBoost and LightGBM, this model is superior to these two single models.