A hybrid machine learning framework for multi-objective performance optimization and anomaly detection in maritime operations

Abstract

This paper introduces a novel hybrid machine learning framework for optimizing key performance indicators in maritime operations. Using real voyage data comprising 1,170 unique voyage records, our methodology integrates a Multi-Task Learning (MTL) model to simultaneously predict vessel efficiency, cost, and turnaround time, capturing their inherent correlations. We then use causal inference to provide prescriptive analytics, estimating the impact of operational decisions like speed changes. An anomaly detection model is also included to identify potential mechanical issues or data errors. Our findings demonstrate that the framework provides a more robust representation of complex maritime trade-offs. The causal analysis quantifies these trade-offs, revealing an average 8.5% cost reduction per knot decrease in speed under optimal conditions. This holistic framework serves as a powerful decision-support tool, helping ship operators enhance both economic and environmental performance.