<p>Accurate prediction of solar activity and geomagnetic disturbances is essential for reducing the risks posed by space weather to technological systems. This study presents a hybrid deep learning model that integrates Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks for simultaneous forecasting of sunspot numbers and geomagnetic indices (Ap and Kp). Using comprehensive datasets spanning solar cycles 20 to 24 (1964–2016), the model employs advanced preprocessing techniques including Savitzky-Golay filtering and normalization and a 60-day sliding window to capture complex temporal dependencies. Model performance, evaluated via 8-fold cross-validation, demonstrates high predictive accuracy, achieving R² values of 0.9943 for sunspot numbers, 0.970 for Kp, and 0.923 for Ap, with low RMSE values. Heatmaps highlighted low RMSE across most time segments, confirming model robustness. Our results confirm a strong correlation between high Ap-index values and increased sunspot activity. Extreme event analysis demonstrates reliable detection of high-intensity geomagnetic storms, with substantial improvements in probability of detection and false alarm rates relative to NOAA/SWPC benchmarks. Comparative assessments show that the hybrid LSTM-GRU model outperforms standalone deep learning and conventional approaches, offering both aggregate skill and operationally relevant performance, even in the presence of severe class imbalance. The hybrid LSTM–GRU model demonstrates clear advantages over standalone LSTM and GRU architectures, with correlation analysis confirming strong links between sunspot activity and the Ap-index, underscoring the model’s ability to capture solar-terrestrial interactions. The proposed LSTM-GRU model demonstrates significant potential for real time space weather forecasting and offers a scalable framework for extended solar-terrestrial predictive analysis.</p>