Empowering econometric methods with machine learning for policy making: A comparative study in maritime transportation

Abstract

The maritime transportation plays a critical role in global trade, yet ensuring its safety and regulatory compliance remains a significant challenge. This study investigates the comparative strengths and limitations of econometric methods and machine learning in supporting policymaking. Leveraging publicly accessible data from port state control (PSC) inspections as the case study, we develop models to identify key factors influencing ship deficiencies (i.e., non-compliance) in PSC and to predict the number of deficiencies during inspections. The results show that machine learning outperforms econometric methods in predictive performance, while econometric methods offer unique advantages in providing interpretable causal insights, enabling a deep understanding of the factors influencing ship deficiencies. Furthermore, by integrating machine learning techniques into econometric frameworks, we uncover nuanced relationships — such as the heterogeneous impact of ship age on ship deficiencies and a U-shaped relationship between ship tonnage and deficiencies — while also enhancing the predictive reliability of econometric methods. By combining the interpretability of econometric methods with the predictive power of machine learning, this study establishes a robust framework for assessing ship risk, enhancing maritime safety management, mitigating maritime risks, and improving transportation policies and regulations.