Repeat Buyer Prediction For E-commerce Based On Machine Learning

With the advent of the Internet era and the popularity of smart phones and computers,the whole society e-commerce application awareness gradually increased,making e-commerce development gradually into a new stage of intensive innovation and rapid expansion.And it has brought much competitive pressure to the existing e-commerce,the major e-commerce platform is particularly fierce competition.In order to compete for the market,the major e-commerce platform businessmen to diversify the promotional activities to attract new customers.At the same time,with the development of information technology,more and more e-commerce platform began to collect customer-based data,because the cost of maintaining the old customers is much smaller than the development of new customers.In order to conduct targeted marketing activities to potential repeat customers,how to use the customer's behavior data to predict which of the new customers in the business during the promotion will become a repeat buyer caused the researchers note.This paper proposes a machine learning algorithm that uses model combination to predict the repeat purchase of customers.First of all,according to the Tmall "double 11" on some of the new business customers and before the half year of the customer's behavior data,extract the relevant features of customers and merchants;Secondly,on the features vector training logic regression,GBM and XGBoost single model;Finally,the prediction results of the optimal single model are weighted by the mixed method to further improve the prediction effect of the model.The work of this paper mainly includes the following aspects:(1)Design feature engineering.By analyzing the factors that affect the repeated purchase behavior of the customer,the basic statistics features,integration features,complex features,age and gender features and recent activity features are proposed,and XGBoost is used to evaluate and select the importance of the features to improve the generalization ability of the models.(2)Research the application of a single model in feature engineering.The linear model logistic regression and the nonlinear model GBM and XGBoost based on the decision tree are selected respectively.After the parameter adjustment of the model,the nonlinear model can better predict the result by using the feature.(3)Research the integration of a single model.In order to avoid underfitting of the linear model,the decision tree model is easy to overfitting,the model combination algorithm is used to integrate the prediction results of the single model,and the predicted results are further improved compared with the single models.