Xie Lun, Liu Jiaquan, Wang Wei
School of Journalism and Communication, Shanghai University, Shanghai, 200072, China.
Anhui Business and Technology College, Wuhu, 231131, Anhui, China.
Sci Rep. 2024 Jun 10;14(1):13297. doi: 10.1038/s41598-024-62368-6.
E-commerce provides a large selection of goods for sale and purchase, which promotes regular transactions and commodity flows. Efficient distribution of goods and precise estimation of customer wants are essential for cost reduction. In order to improve supply chain efficiency in the context of cross-border e-commerce, this article combines machine learning approaches with the Internet of Things. The suggested approach consists of two main stages. Order prediction is done in the first step to determine how many orders each merchant is expected to get in the future. In the second phase, allocation operations are conducted and resources required for each retailer are supplied depending on their needs and inventory, taking into account each store's inventory as well as the anticipated sales level. This suggested approach makes use of a weighted mixture of neural networks to anticipate sales orders. The Capuchin Search Algorithm (CapSA) is used in this weighted combination to concurrently enhance the learning and ensemble performance of models. This indicates that an effort is made to reduce the local error of the learning model at the model level via model weight adjustments and neural network configuration. To guarantee more accurate output from the ensemble model, the best weight for each individual component is found at the ensemble model level using the CapSA method. This method yields the ensemble model's final output in the form of weighted averages by choosing suitable weight values. With a Root Mean Squared Error of 2.27, the suggested technique has successfully predicted sales based on the acquired findings, showing a minimum decrease of 2.4 in comparison to the comparing methodologies. Additionally, the suggested method's strong performance is shown by the fact that it was able to minimize the Mean Absolute Percentage Error by 14.67 when compared to other comparison approaches.
电子商务提供了大量可供买卖的商品，促进了常规交易和商品流通。高效的商品配送和对客户需求的精确预估对于降低成本至关重要。为了在跨境电子商务背景下提高供应链效率，本文将机器学习方法与物联网相结合。所提出的方法包括两个主要阶段。第一步进行订单预测，以确定每个商家未来预计会收到多少订单。在第二阶段，进行分配操作，并根据每个零售商的需求和库存，考虑每个店铺的库存以及预期销售水平，为其提供所需资源。所提出的方法利用神经网络的加权混合来预测销售订单。在此加权组合中使用卷尾猴搜索算法（CapSA）来同时提高模型的学习和集成性能。这表明通过模型权重调整和神经网络配置，在模型层面努力减少学习模型的局部误差。为了确保集成模型输出更准确，使用CapSA方法在集成模型层面找到每个单独组件的最佳权重。该方法通过选择合适的权重值，以加权平均值的形式产生集成模型的最终输出。根据所得结果，所提出的技术成功预测了销售额，均方根误差为2.27，与比较方法相比，最低下降了2.4。此外，与其他比较方法相比，所提出的方法能够将平均绝对百分比误差最小化14.67，这表明了该方法的强大性能。
