Sales prediction-driven dynamic selection of logistics modes for cross-border e-commerce considering products return

With the continuous growth of cross-border e-commerce and the increasing volatility in global demand and logistics costs, cross-border e-commerce enterprises face significant challenges in dynamically selecting logistics modes and coordinating inventory and transportation decisions. To address these challenges, this research proposes an integrated optimization framework for cross-border e-commerce enterprises with a self-built logistics system, aiming to dynamically select between Bonded Warehouse and Direct-Mail modes across multi-product, multi-period, and multi-destination, while jointly optimizing warehouse allocation and cross-border trunk transportation. To improve demand forecasting accuracy, an attention-augmented sequence-to-sequence long short-term memory framework is developed. To better consider the impact of return behavior on the selection of logistics mode, we establish a method to quantify the product return rate based on customer utility. Based on the predicted demand and product return rates, a multi-period, multi-product, and multi-destination optimization model is formulated to minimize logistics costs. To solve this NP-hard problem, a hybrid heuristic algorithm is proposed. A real-world case study involving Hong Kong, Macau and Taiwan markets validates the effectiveness of the proposed method. Experimental results reveal key insights into logistics mode selection under varying cost structures, return behaviors, and regional characteristics, offering practical guidance for responsive and cost-efficient cross-border e-commerce logistics strategies.