A data-driven geographic information system and machine learning based multi-criteria framework for strategic wind power plant siting

Abstract Wind energy site selection requires robust frameworks that simultaneously address expert uncertainty, objective efficiency screening, and predictive capability beyond sampled locations. This study presents an integrated framework for strategic wind farm site selection in Ethiopia’s Amhara Region by combining Fuzzy Analytic Hierarchy Process (FAHP), efficiency analysis, and predictive modelling to overcome the limitations of static GIS approaches. The FAHP stage incorporates expert judgment through fuzzy triangular numbers to weight six variables such as wind speed, slope, elevation, distance to transmission lines, distance to roads, and land-use/land-cover to achieve a consistency ratio of 0.0175 with wind speed emerging as the dominant factor of 0.4211 weights in order to generate a spatial suitability surface and extract candidate high-potential areas. High-potential zones identified by FAHP are then evaluated as decision-making units using input-oriented Data Envelopment Analysis (DEA) models. DEA efficiency screening identifies North Shewa as frontier-efficient zone (CCR = BCC = SBM = 1.0), which contributes 32.25% of regional suitable land. Finally, machine learning (ML) models (Random Forest (RF), Support Vector Machine (SVM) and Extreme Gradient Boost (XGBoost)) are trained on 1.5 million sampled pixels from the 31-million-cell feature space to learn generalizable suitability functions. Random Forest achieved optimal performance with RMSE of 0.2981 and R 2 of 0.8145 in regression and F1-score of 0.9207 and accuracy of 0.9237 in classification to delineates 1,698 km 2 of high-priority corridors within North Shewa for immediate wind farm deployment. Independent validation through five different sources that includes 250 MW of Debre Birhan under Ethiopian Ministry of Finance private public project pipeline to confirm North Shewa as the highest-potential development corridor. This framework advances wind energy planning by integrating subjective weighting, objective efficiency analysis, and predictive modelling into a unified strategic decision-support system applicable for wind energy planning in data-scarce region.