This study addresses a critical gap in the literature by integrating big data and dynamic optimization to overcome the limitations of traditional Multi-Criteria Decision-Making (MCDM) methods in shipyard site selection in Istanbul. Key factors such as environment, logistics, security, and energy infrastructure were evaluated using Machine Learning (ML) algorithms, including Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), and Categorical Boosting (CB). Among them, RF (F1 score 0. 98) and CB (F1 score 0. 93) demonstrated the highest predictive performance. The weights derived from these models were optimized through Game Theory (GT) based on the Nash Equilibrium to minimize bias. GIS-based spatial analysis identified sea depth, power transmission lines, slope, and elevation as decisive variables. Suitability classification revealed that 49. 94% of the area was highly suitable (Class 4–5), 18. 68% moderately suitable (Class 3), and 31. 39% of low suitability (Class 1–2). The fact that existing major shipyards in Tuzla and Pendik fall within highly suitable zones validates the model's reliability. Istanbul was divided into 29 candidate regions; initial ML-based rankings and decision map-based rankings were reconciled through GT. As a result, AZ₁8, AZ₂5, and AZ₂4 were identified as the most suitable areas for future shipyard development. • ML-GIS integration enables objective, data-driven shipyard site selection. • RF (F1 = 0. 98) and CB (F1 = 0. 93) showed highest predictive accuracy. • Game Theory reduced model bias using Nash Equilibrium optimization. • 49. 94% of study area classified as highly suitable for shipyard sites. • AZ₁8, AZ₂5, AZ₂4 identified as top-priority zones for investment.
Doğan et al. (Wed,) studied this question.