The Intensive Care Unit (ICU) is dedicated to delivering critical care for patients with severe medical conditions, emphasizing the importance of accurate early mortality prediction in effective patient management. Early prediction of ICU mortality enables timely medical interventions and efficient resource allocation. This work introduces a robust prediction framework employing deep learning models, specifically the LSTM and RNN architectures, to predict patient mortality using clinical data. The framework addresses class imbalance using RandomOverSampler and leverages the strengths of sequential data modeling. The LSTM model achieved exceptional results, including an AUC-ROC of 0.9744, precision of 0.8608, recall of 0.9763, F1-Score of 0.9149, training accuracy of 93.85%, and validation accuracy of 92.99%. The RNN model also demonstrated effective performance, achieving an AUC-ROC of 0.9293, precision of 0.8102, recall of 0.8157, F1-Score of 0.8129, training accuracy of 84.70%, and validation accuracy of 85.91%. These findings highlight the superior predictive power of the LSTM model, while also demonstrating the effectiveness of both models in analyzing temporal clinical data. This approach provides a powerful tool for early intervention and efficient ICU resource management. Future work will focus on incorporating additional patient characteristics, improving the model explainability, and validating the framework in various clinical settings to improve its applicability and generalizability.