Dynamic demand and inventory prediction model for cross-border E-commerce based on ARIMA and LSTM

Abstract

The traditional cross-border e-commerce demand and inventory prediction model has poor real-time and efficiency, and is difficult to undertake complex tasks. To address the above problems, the study proposes a dynamic demand and inventory prediction model for cross-border e-commerce based on differential autoregressive integrated moving average model and long short-term memory. This can optimize the timing of demand prediction and reduce the cost of inventory management, and finally set up experiments to verify it. The experimental results show that in the simulation running experiments, the average demand forecasting efficiency of the proposed model before training is 50.70%, and it is improved to 97.56% after training; the average inventory control accuracy of the model after training reaches 0.94, and the average deviation of accuracy is 0.009. After the model loses the demand forecasting module the demand forecasting efficiency decreases to 84.43%, and the inventory control accuracy decreases to 0.84. After the model loses the inventory management module the average demand forecasting efficiency decreases to 93.86%, and the inventory control accuracy decreases to 0.933. After losing the inventory management module, the demand forecasting efficiency drops to 93.86%, and the inventory control accuracy drops to 0.933. In the actual model performance experiments, the average demand forecasting efficiency in the sales smooth phase is 90.02%, and the inventory control accuracy is 0.90; the two indexes in the non-smooth phase are 81.15% and 0.82, respectively. The trajectory of the demand coefficient predicted by the model is 0.009 (the lowest value among all compared methods). The study can improve the supply chain demand and inventory prediction efficiency and real-time, and enhance the market competitiveness of the cross-border e-commerce in a larger field.