Deep Learning-based Multivariate LSTM Approach for Forecasting Export and Import Container Throughput at Busan Port

Key Points

• The proposed multivariate LSTM reduces mean absolute error (MAE) by 28.7% for exports and 26.6% for imports, proving superior prediction accuracy.
• For mean absolute percentage error (MAPE), the model achieves lower rates of 4.07% for exports and 4.18% for imports than NeuralProphet and SARIMA.
• The analysis incorporates external variables like exchange rates and industrial production indices to enhance prediction accuracy.
• Findings suggest significant improvements in forecasting capabilities for container throughput, highlighting the model's effectiveness.

Abstract