Research Information System
Ocean Engineering, vol.344, 2026 (SCI-Expanded, Scopus)
This study presents the first offshore wind-farm siting model in the Eastern Black Sea integrating a fully data-driven Unsupervised Machine Learning–Geographic Information System (UML–GIS) framework. The model ensembles four unsupervised algorithms—Isolation Forest (IF), Unsupervised Random Forest (URF), k-means (KM), and Gaussian Mixture Model (GMM)—to generate a composite suitability ranking validated through bootstrap and Leave-One-Criterion-Out (LOCO) analyses. Fifteen multidisciplinary criteria encompassing physical, environmental, and socio-economic dimensions were evaluated. Principal Component Analysis (PCA) preserved 93.6 % of total variance, confirming low redundancy among variables. The ensemble exhibited strong consistency with IF (ρ = 0.82) and URF (ρ = 0.87), while bootstrap analysis identified ten highly stable core zones (stability = 0.95–1.00) across 1000 iterations. LOCO validation confirmed that no single criterion dominated the results (ρ = 0.96–1.00). Comparative analysis with the classical Multi-Criteria Decision-Making (MCDM) method, Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), showed a significant inverse correlation (ρ = −0.633, p < 0.001), indicating that the UML–GIS framework captures complex, non-linear inter-criteria relationships beyond traditional linear models. The framework yields reproducible, unbiased, and statistically robust spatial decisions, offering a transparent and policy-ready tool for sustainable offshore wind planning.