This study investigates the application of machine learning models to address after-sales service issues in cross-border e-commerce, focusing on predicting order returns to reduce return costs and optimize customer experience. Using H cross-border e-commerce company as a case study, the research employs Random Forest and XGBoost models to identify high-risk return orders. By comparing the performance of these two models, the study highlights their respective strengths and weaknesses and proposes optimization strategies. The findings provide a valuable reference for e-commerce companies to refine their business models, reduce return rates, improve operational efficiency, and enhance customer satisfaction.