Research on E-commerce User Purchase Behavior Prediction Based on Machine Learning

Abstract

E-commerce platform is getting bigger and bigger, which can gather a lot of user behavior data. This gives you a new opportunity to know how consumer buy things. By analyzing these data, enterprises can improve the efficiency of marketing and provide users with services that are more suitable for everyone. In this study, the user activity records collected from e-commerce platform were used to investigate the prediction of whether to buy or not. Turn the information of browsing, operating shopping cart, using coupon, buying before, and these behaviors into the expected features. Using the algorithm of Logistic Regression, Decision Tree and Random Forest, a prediction model is established to evaluate the user’s “mood of wanting to buy”. In order to see the effect of the model, a comparative experiment was carried out under the same training and testing conditions. The experimental results show that the overall prediction accuracy rate is the highest in the dataset selected by Logistic Regression. The results show that the behavior variables used in this study have clear patterns, and even the relatively simple linear model can be well captured. Random Forest has a more complex model structure, but the performance is not obvious according to the current data size and feature settings. The research shows that it is better to choose an algorithm that conforms to the characteristics of dataset than to rely only on the complexity of the model. This result may be helpful for purchasing forecasting and precision marketing application in e-commerce environment.