Research on Prediction of E-commerce Users’ Purchase Behavior Based on Gradient Boosting Decision Tree Model

Driven by big data, e-commerce platforms have accumulated massive user behavior data, which can be transformed into valuable information after cleaning and feature selection. This study analyzes users’ historical behavioral data on e-commerce platforms, constructs a gradient boosting decision tree prediction model based on user, product, category, and two-by-two interaction behavioral features, and extracts designed feature data from raw CSV data based on Hiv as the prediction basis of the model. At the same time, clustering analysis is performed based on the user’s purchasing behavior (dwell time, browsing frequency) to generate user profiles. The experimental results show that after 7 days, the purchase conversion rate of browsing, collecting, adding to cart and purchasing tends to 0. Therefore, the time window for purchase behavior prediction is chosen to be 7 days. In this paper, the prediction model is only trained to 20 epochs, and the Loss value converges to about 0.14, which shows a good training effect. The model has the best classification performance for user purchase behavior prediction, with precision, recall, and F1 values between 0.91 and 0.97. The clustering algorithm divides the user purchase behavior into four clusters, where cluster class 4 has the best user value. In summary, using the gradient boosting decision tree model, e-commerce platforms can more accurately predict user purchasing behavior, thus improving user experience and platform economic benefits.