Abstract：

In the rapid development of digital economy, the logistics industry is transitioning from laborintensive to technology-intensive. Accurate logistics demand forecasting is essential for supply chain operations and strategic planning. But traditional forecasting methods can't grasp the complicated, non-linear features of logistics demand in the big data time. This paper builds an accurate forecasting model by integrating LSTM networks and multi-dimensional economic indicators. We aggregate datasets such as regional GDP, retail sales, historical freight volumes to build up a full set of feature engineering. An empirical analysis of a representative economic zone is carried out showing the results of the proposed a novel big data driven model performs much better than the conventional algorithms, both in terms of higher precision as well as in terms of better stability. From the results we can say that incorporating external macroeconomics factors with internal data using advanced machine learning can reduce the supply chain “bullwhip effect”. Therefore, this study provides both theoretical and practical insights for logistics organizations to optimize resource allocation in volatile markets.

Key Words：

logistics demand prediction; big data analytics; Long Short-Term Memory (LSTM); empirical analysis; supply Chain optimization