Repurchase Prediction of Community Group Purchase Users Based on Stacking Integrated Learning

Abstract

Recently, fewer scholars consider the prediction of repeat purchases in new retail models. Based on the real data of community group buying enterprises, this paper will study the prediction of community group buying users' repurchase behavior. Firstly, this paper carries out feature engineering according to the characteristics of the community groups buying industry. Finally, 313 features are extracted from the user dimension, head dimension, and business personnel dimension, respectively. Then, based on the heterogeneous integrated learning method stacking, three two-tier fusion models with the same primary learners but different secondary learners are constructed. Two homogeneous ensemble learning models, random forest and lightgbm, and the traditional single machine learning model are introduced for comparative experiments. Experiments show that the fusion model based on ensemble learning method has better prediction performance than a single model. Among the fusion models, the stacking two-layer fusion model with neural network model as secondary learner is the best.