Autoformer-Based Sales and Inventory Forecasting for Cross-Border E-Commerce: A Time Series Deep Learning Approach

Accurate forecasting of product sales and inventory is a critical task for cross-border e-commerce platforms, where demand volatility, long logistics cycles, and dynamic pricing present significant challenges for efficient supply chain management. Traditional statistical and short-term machine learning models often fail to capture long-term dependencies and complex seasonal variations in sales data, leading to inaccurate demand planning and inefficient inventory allocation. To address this limitation, we propose a forecasting framework based on the Autoformer model, a deep learning architecture designed for long-sequence time series prediction. The model leverages series decomposition blocks to separate trend and seasonal components, and employs an auto-correlation mechanism to enhance the capture of periodic demand patterns in product sales and inventory turnover. We evaluate the framework on a real-world cross-border e-commerce dataset comprising transaction-level order volume, prices, inventory records, and external market indicators such as exchange rates and shipping costs. Experimental results show that the proposed Autoformer-based approach achieves superior forecasting accuracy compared with baseline models including LSTM, Transformer, and Informer. Specifically, our model reduces prediction error with a Mean Absolute Error (MAE) of 18.6 and a Root Mean Square Error (RMSE) of 25.4, representing a 17.3% improvement over the best-performing baseline. These findings highlight the potential of Autoformer for enhancing sales forecasting, reducing stockouts, and improving inventory turnover in cross-border e-commerce platforms, thereby supporting more effective logistics management and strategic decision-making.