Machine Learning Analysis of Coastal Water Pollution in China: Drivers and Complex Relationships

This study investigates the nonlinear effects of socioeconomic, pollution control, and technological innovation factors on seawater pollution using a Generalised Additive Model (GAM) and panel data from 11 Chinese coastal provinces (2018-2023). Key findings indicate: (1) Chemical Oxygen Demand Emissions (CODE) from direct marine discharge sources showed a temporal pattern of “initial increase, subsequent decrease, and gradual stabilisation”, with significant spatial heterogeneity - Zhejiang recorded the highest emissions, Tianjin the lowest; (2) GAM revealed significant nonlinear relationships between most variables and CODE; (3) Factor impacts exhibited distinct range-dependent characteristics. These findings provide a scientific basis for identifying key pollution sources and formulating differentiated control strategies, leading to targeted policy recommendations.