Comparative Review of Machine Learning Models for Sunspot Number Prediction

Abstract

Accurate prediction of sunspot numbers is essential for understanding solar activity and mitigating the adverse effects of space weather on technological infrastructure. With the limitations of traditional statistical methods, recent years have witnessed a surge in the application of machine learning (ML) models, particularly deep learning architectures, to sunspot time series forecasting. This review presents a comprehensive comparative analysis of major ML models utilised in the prediction of sunspot numbers, focusing on recurrent neural networks (RNN), long short-term memory (LSTM) networks, gated recurrent unit (GRU) models, and hybrid neural network approaches. The article synthesises findings from state-of-the-art literature, summarising the methodological advances, dataset preparation strategies, and evaluation metrics commonly employed in this field. A critical assessment of model performance, based on accuracy, robustness, and operational feasibility, highlights the superior capabilities of LSTM and GRU architectures for long-term and multi-step forecasting tasks. By systematically evaluating methodological advancements and benchmarking results from recent studies, this article highlights the strengths, limitations, and emerging trends in solar forecasting approaches, aiming to guide future research toward robust, interpretable, and operationally feasible sunspot prediction.