Customer churn prediction analysis on E-commerce platforms based on improved XGBoost algorithm

Abstract

Customer churn prediction and analysis on e-commerce platforms are key to e-commerce systems, providing strong support for e-commerce and traffic planning. As the advancement and application of deep learning, e-commerce platform customer churn prediction models with traditional XGBoost algorithms have emerged one after another. However, traditional XGBoost algorithms have many drawbacks and limitations. These problems with slow training speed, accuracy loss, overfitting, and inability to handle large-scale data. Therefore, an improved XGBoost algorithm is introduced to improve accuracy under the influence of data mining and minimal training resources. Experimental results demonstrate that the proposed model achieves an accuracy of 78.7%, a prediction precision of 73.5%, and a recall rate of 58.7%. For core and important-development customer categories, the AUC values reached 86.7% and 86.5%, respectively. Compared to baseline models such as SVM, BP neural networks, and C4.5 decision trees, the improved XGBoost demonstrates superior accuracy and stability. These findings confirm that the predicted values are closer to the true labels and that the method provides more reliable and accurate support for customer segmentation and retention strategy development on e-commerce platforms.