A resilient model for trade volume forecasting under economic uncertainty: Addressing challenges in the global supply chain

In recent years, the escalating economic uncertainty arising from Sino-US trade frictions has made the accurate forecasting of trade volume a crucial yet challenging task, with wide-ranging implications for the stability of the global supply chain. Precise trade forecasts are essential for supporting strategic decision-making and ensuring resilience across industries that rely on international trade. To address this challenge, this study introduces an innovative predictive model, the principal component analysis-simulated annealing-backpropagation neural network (PCA-SA-BPNN), specifically developed to enhance forecasting accuracy within this volatile economic landscape. The model utilizes principal component analysis (PCA) to reduce the dimensionality of extensive datasets collected from search engines, simplifying the data while retaining critical information. Simultaneously, simulated annealing (SA) is applied to optimize the backpropagation neural network (BPNN), effectively addressing the local optimization challenges often impair traditional backpropagation neural network models, which can hinder prediction accuracy. The effectiveness of the PCA-SA-BPNN model is demonstrated through comprehensive comparative experiments, demonstrating its superior performance compared to other models, including principal component analysis-adaptive differential evolution-backpropagation neural network (PCA-ADE-BPNN) and principal component analysis-backpropagation neural network (PCA-BPNN) models, as well as standalone XGBoost and BPNN models. The PCA-SA-BPNN model achieves notably lower mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values, with an R2 approaching 1, underscoring its superior predictive performance. This research thus offers valuable insights into how combining dimensionality reduction, optimization techniques, and neural network networks can significantly enhance trade volume forecasting amidst economic uncertainties. Furthermore, it provides valuable insights into the interplay between predictive accuracy, model efficiency, and resilient decision-making within global supply chain management, contributing to both theoretical advancements and practical applications in the field.