User online purchase behavior prediction based on fusion model of CatBoost and Logit

Abstract

Analyzing users' consumption preferences and willingness based on massive online behavior data can help optimize user experience and improve product purchase rate. For the online platform user data, feature engineering methods were used to pre-process, visualize, analyze and construct features. Then, a prediction and evaluation model of users' online purchase behavior was developed based on the CatBoost algorithm, and the important features affecting users' purchase behavior were quantitatively analyzed using the Logit Model. Compared with Random Forest, Support Vector Machine and XGBoost, the prediction model has better performance with prediction accuracy of 98.38% and F1 score of 0.775. The results show that user basic information, user access data and user login data have impact on user online purchase behavior. Especially the number of visits with or without coupons, unfinished orders of annual classes, number of coupons received, repeated learning of courses, public number following, days of login, city and number of learning classes sections are important features affecting purchase behavior.