Global renewable energy forecasting using hybrid ML/DL models: Economic and geospatial insights

This study develops a globally applicable renewable energy forecasting framework that integrates machine learning (ML), deep learning (DL), and statistical methods, and calibrates them to economic and regional conditions. Using data for more than 200 countries, sequence models perform best in data-rich cohorts: GRU and LSTM attains the lowest MAPE in North America (0.0210) and Oceania (0.0699). In income-stratified regressions, Random Forest achieves R2-score of 0.9153 for high-income countries, whereas lower-middle-income cohorts exhibit higher errors. G20 countries lead development (1700 TWh wind), and the global mix has shifted from historical hydropower dominance toward wind and solar as costs decline and technologies diffuse. This study provides valuable insights for policymakers and stakeholders, promoting the development of effective energy policies and strategies to enhance the global transition to sustainable energy systems. Policy-relevant implications follow: in data-scarce settings, prioritize grid digitalization and standardized reporting before adopting complex DL; in data-rich settings, favor GRU/LSTM with rolling retraining and uncertainty bands. Future research should benchmark transformer-based and hybrid pipelines and design capacity-building strategies for lower-income countries.