Ensembling XGBoost and Neural Network for Churn Prediction with Relabeling and Data Augmentation


This paper describes our solution in KKBOX’s Churn Prediction Challenge, one of tasks in WSDM Cup 2018. The competition aims at predicting whether the KKBOX’s users will churn after a period of time. To build a competitive system, we first enrich training set by data augmentation and relabeling, and then carefully design specific features for this problem. By ensembling the models of neural networks and XGBoost, our team PKU Fresher ranks 6th among 575 teams on the final private board.

In International Conference on Web Search and Data Mining Workshops (WSDM Workshop).