Coronal holes significantly impact space weather, making understanding and predicting their activity crucial for mitigating effects on Earth’s technological infrastructure. This paper presents a novel approach to forecasting short-term coronal hole activity using a multi-head CNN-LSTM model trained on Solar Dynamics Observatory (SDO) dataset tabular data and images. Our methodology involves preprocessing, training, and implementing a prediction system to predict the solar surface ratio covered by coronal holes. We compare our results with baseline models, including LSTM, GRU, and ARIMA-based models, demonstrating improved predictive performance. Sensitivity studies analyze parameter impacts on performance. Our work contributes to filling the gap in predictive modeling for coronal hole activity, providing valuable insights for space weather forecasting and preparedness. The RESNET-50 based LSTM-CNN model outperforms comparable models by 34% in MSE and 27% in MAE in predicting sunspot ratio, identifying optimal look back and predict ahead windows that further improve performance by up to 15%. Future work includes extending the model to longer-term predictions and incorporating additional features for enhanced accuracy.