A comparative assessment of spatial regression methods for predicting maritime events in the Caribbean Sea

Abstract

Predicting maritime events requires spatially aware techniques that account for complex geographic and operational dynamics. By applying Geographic Weighted Regression, Generalized Linear Regression, Forest-Based Classification and Regression, and Empirical Bayesian Kriging, this research models the spatial distribution of vessel events across the Caribbean Sea. Key influencing variables include vessel traffic density, charted zone confidence, flag state, and vessel age. Results highlight regional hotspots such as the Panama Canal and Gulf of Paria. Comparative analysis demonstrates the strengths and limitations of each technique, informing the development of adaptive, location-specific maritime risk mitigation strategies.