Evaluating AI approaches for space weather prediction: Strengthening satellite resilience and technology systems

The growing reliance on satellite communications, navigation systems, and power grids has amplified the need for accurate and timely space weather prediction. Traditional empirical and physics-based models, while foundational, remain computationally expensive and often struggle with extreme or rare events. In this study, we present a structured comparative framework that systematically evaluates machine learning (ML) and deep learning (DL) methods for forecasting diverse space weather phenomena, including solar flares, coronal mass ejections, geomagnetic storms, and electron flux. Unlike prior reviews, this work contributes (i) a taxonomy of AI algorithms mapped to specific space weather applications and (ii) a synthesis of comparative performance across datasets and metrics such as mean absolute error, RMSE, and false alarm rate. By consolidating fragmented literature and benchmarking algorithms across multiple event types, this paper provides a holistic perspective on how AI can advance the resilience of space-based and terrestrial technologies. Our findings indicate that while tree-based methods (e.g., Random Forest, XGBoost) excel at short-term geomagnetic forecasts, deep learning approaches (CNNs, LSTMs) offer superior performance for image-based and temporal predictions, and hybrid models show promise for real-time operational forecasting. This synthesis serves as both a reference and a guide for future research aimed at developing robust, scalable, and interpretable AI-driven systems for space weather prediction.