Predicting the mortality state of each patient in Intensive Care Unit is important to any stakeholder so that his/her care and treatment can be appropriately prepared and changed in practice. Such importance is reflected via many existing works which were dedicated to the task from many different perspectives. However, data imbalance handling has not yet been examined thoroughly in a general manner. Focus on true “dead” patients in model construction has not yet been discussed. These two issues might hinder the task resolution and applicability in practice. Therefore, in this paper, we propose an imbalance-aware ensemble model to include data imbalance handling at the core of an ensemble learning process in favor of the minority class of “dead” patients. The proposed model is then effectively generated in the bagging mechanism. As a result, our model achieves the highest F1-Score and AUC in all the experiments on the popular MIMIC III database in comparison with several other typical models previously used in the existing works. Such a result confirms the capability of our model to predict more true “dead” patients correctly and thus, the model can be embedded in practical applications.