Forecasting Container Throughput in Northwest Europe’s Inland Waterways: A Data-Driven Approach

Abstract

Accurate forecasting of container throughput volume (CTV), measured in Twenty-foot Equivalent Units (TEUs), is important for optimizing logistical planning and resource allocation in inland waterway transportation (IWT) systems. Failure to anticipate fluctuations in container volumes may lead to underutilization of resources and overwhelming congestion, resulting in significant financial losses. The literature shows the limitation of traditional individual CTV prediction models, highlights hybrid models' promise, and identifies a research gap in inland waterway CTV forecasting for Northwest Europe covering Belgium, Germany, Netherlands, and France. This study addresses this gap by employing a hybrid machine learning approach that integrates Artificial Neural Networks (ANN), Gated Recurrent Units (GRU), and Bayesian Neural Networks (BNN) with Support Vector Regression (SVR), enhanced by Random Forest-Recursive Feature Elimination (RF-RFE). Collected Eurostat (2007-2023) data includes economic, environmental, and logistical variables relevant to CTV. Results show that the hybrid model outperforms individual models, achieving the lowest error rates across all countries. In Belgium, it reached a Root Mean Squared Error (RMSE) of 12,661.12 TEU and a Mean Absolute Percentage Error (MAPE) of 2.29%. Similarly, it yielded strong results in Germany (RMSE 20,066.66 TEU, MAPE 2.81%), Netherlands (RMSE 31,593.75 TEU, MAPE 2.22%), and France (RMSE 9,164.19 TEU, MAPE 4.11%). These results affirm the hybrid model’s effectiveness in enhancing CTV forecasting accuracy, aiding stakeholders in making informed decisions, reducing congestion, and improving IWT reliability. Future work will focus on real-time forecasting with intermodal transport systems, and assessing the environmental impacts of predicted freight patterns.