Enhancing cross-border forest product trade forecasting with LSTM and multi-source data fusion

Accurate demand forecasting for cross border forest products is essential for managing trade and mitigating supply chain disruptions. Traditional methods like ARIMA and SARIMA often fail to capture the complex, multi dimensional relationships that affect demand. This paper proposes an advanced LSTM model integrated with multi source data fusion and a dynamic attention mechanism to address these challenges. By combining data from diverse sources economic indicators, climatic conditions, market trends, and geopolitical factors our model improves forecasting accuracy. The dynamic attention mechanism adapts to varying conditions, prioritizing the most relevant data at different times. Experimental results demonstrate that our model outperforms traditional methods and other machine learning baselines, achieving a Mean Absolute Percentage Error (MAPE) of 10.1%, a Root Mean Squared Error (RMSE) of 0.150, and an R $$^2$$ value of 0.90. These findings underscore the potential of multi-source data fusion and dynamic attention in improving demand forecasting in the forest product trade.