Online Shopping Customer Churn Prediction Based On Stacking Ensemble Learning

In recent years,with the rise of the e-commerce model,online shopping has become increasingly popular.While online shopping brings a good value experience to consumers,the instability of platform user behavior increases and the rate of customer churn is higher.The creation of new customers will bring benefits to the enterprise,but comparatively speaking,the retention of old customers requires less cost and greater profits.Therefore,if e-commerce enterprises want to make themselves stand out in the fierce market competition and occupy the advantages,they must pay attention to the effective management of customers and reduce the customer churn rate.Online shopping customers are different from those in the telecommunications and banking industries.Customers in the telecommunications and banking industries are typical contract customers.The customer status is easy to define,while online shopping customers,such as e-commerce website customers,are non-contract customers.Instability,customer status is difficult to predict.In the research of customer churn prediction at home and abroad,the industries such as telecommunication and banking are involved in more,and the research results are rich.Compared with the industries such as telecommunication and banking,the research results of online shopping customer churn prediction at home and abroad are less,and the prediction methods in the existing research mainly use a single learner and improve a single learner.In view of the above problems,this paper uses stacking ensemble learning method to build the online shopping customer churn prediction model,and conducts experiments on the prediction effect of the model on different online shopping customer data sets.In the process,the coverage weight and K-means clustering are integrated into the research,which has a certain theoretical and practical significance.The main research work is as follows: First,churn prediction based on weighted Stacking ensemble learning.The first mock exam is based on customer data of J e-commerce website.First of all,a single model is used to predict customer churn.On this basis,a two level fusion model of Stacking ensemble learning is built.During the combination of two level data sets,the coverage weight is innovatively integrated.The weight of the prediction coverage rate of different basic learning units is multiplied by the prediction results of the respective base learning devices as the two level data set.Secondly,weighted stacking ensemble learning churn prediction with clustering.Taking KKBOX music streaming media platform online shopping customer data as a research sample to further explore the prediction effect of online shopping customer churn prediction model based on Stacking ensemble learning method.Aiming at the problem that the data is large in volume and severely imbalanced in categories,it cannot be directly used for modeling.First,the pre-processed sample data is clustered based on K-means clustering technology,and then the undersampling method further balance the data set,and finally applied to the two-layer fusion model based on weighted Stacking ensemble learning for churn prediction.The results of relevant empirical studies show that the two-layer fusion model based on weighted Stacking ensemble learning has a significant improvement in hit rate,coverage,accuracy,recall rate and F1 score compared with the single prediction model,and the two-layer data set based on coverage weight has a positive effect on correctly predicting the churn of customers.