E-commerce logistics as an important pillar industry of cross-border e-commerce transactions, how to use digital technology to improve the core competitiveness of the logistics industry chain has become a problem for cross-border e-commerce physical enterprises to think about. In view of the influence of floating factors such as demand, exchange rate and tariff in cross-border e-commerce trade, the way to optimize the cross-border e-commerce logistics chain driven by digital technology based on machine learning is studied. The historical data of cross-border e-commerce logistics overseas warehouse, border warehouse site selection and inventory are used to build a model of cross-border e-commerce random trade scenario, and to optimize the process of logistics warehouse inventory from prediction to decision-making. To test the effectiveness of the models built by applying machine learning techniques for optimising the logistics of actual cross-border e-commerce, a sensitivity analysis is conducted with the cross-border e-commerce of S clothing companies for South-East Asia. Analysis results show that machine learning-based random forest models have an excellent effect on inventory forecasting during the case analysis process, improving the accuracy of cross-border warehouse inventory forecasts, reducing cross-border logistics and e-commerce transit times and promoting the efficiency of cross-border e-commerce logistics services The study found.