Data-driven prediction of recoverable ballast water: implications for port and maritime sustainability

Abstract

To mitigate the potential environmental risks of ballast water discharge and promote water reuse, ports have begun deploying ballast water recovery systems. However, inaccurate forecasting of recoverable ballast water often leads to unnecessary discharge. This study proposes a two-stage classification-regression framework for predicting recoverable ballast water volumes based on ship operation and recovery data. The classification stage effectively addresses data zero-inflation, while the regression stage integrates ridge regression, random forest, and XGBoost through a stacking ensemble strategy to enhance prediction accuracy. Applied to a major dry bulk port in northern China, the proposed method achieves an R² of 0.93. It enables an annual reduction of 300,000 m³ of ballast water discharge, along with decreases of 250 kg of total nitrogen and 190 kg of sulfides. These results demonstrate the effectiveness in reducing environmental risks and enhancing water resource utilization, contributing to sustainable development in maritime transport.