DEEP LEARNING FRAMEWORK USING 2D CNN FOR SUNSPOT-BASED SPACE WEATHER PREDICTION

Abstract

Accurate sunspot prediction is vital for space weather forecasting to protect space-based infrastructure. This study leverages deep learning models—LSTM, GRU, and 2D Convolutional Neural Networks (2D-CNN)—within a cloud computing framework to enhance prediction accuracy and scalability. Using time-series sunspot data, LSTM and GRU capture sequential patterns, while optimization techniques reduce complexity and overfitting. Experimental results show 2D-CNN achieves the highest accuracy at 99.39%, with strong precision and recall, highlighting its superior spatial feature extraction. GRU outperforms LSTM in sequential data handling. These results demonstrate the effectiveness of deep learning, especially 2D-CNNs, for robust sunspot forecasting.