ADVANCING SPACE WEATHER PREDICTION THROUGH DEEP LEARNING: A 2D CNN FRAMEWORK FOR SUNSPOT FORECASTING

Abstract

Accurate prediction of sunspot activity is essential for reliable space weather forecasting and the protection of space-reliant infrastructure. Recent advances in cloud computing have accelerated the adoption of deep learning methods, enabling scalable and efficient predictive models. This study investigates the application of Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Two-Dimensional Convolutional Neural Networks (2D-CNN) for enhancing sunspot forecasting accuracy. Leveraging a cloud-based framework, the approach improves computational efficiency, supports real-time analysis, and facilitates large-scale model deployment.

The dataset comprises time-series records of sunspot activity, making it particularly suitable for recurrent architectures. LSTM and GRU effectively capture sequential dependencies, while optimization strategies such as modified particle swarm optimization and hyperparameter tuning mitigate overfitting and computational overhead. Experimental evaluation demonstrates that 2D-CNN delivers the highest predictive performance, achieving 99.39% accuracy, 99.45% precision, 99.33% recall, and a 98.79% F1-score, underscoring its strength in capturing spatial correlations within sunspot data. Additionally, GRU outperformed LSTM in sequential modeling, with higher precision (98.80% vs. 97.81%) and F1-score (96.21% vs. 96.11%).

Overall, the findings highlight the effectiveness of deep learning—particularly 2D-CNNs—in advancing accurate, scalable, and real-time sunspot forecasting for space weather prediction.