Patients in critical care are at risk of clinical deteriorations such as congestive heart failure, when the heart muscle could not pump the blood as good as it should be, and sudden cardiac arrest, which is caused by the disruption of the heart’s pumping function These clinical deteriorations often appear without any warning. They can cause patients losing consciousness, having clinical complications, or death. A timely treatment is necessary after such complications. The early detection of clinical deteriorations can both improve patient outcomes and reduce the utilization of healthcare resources. Many early warning scores (EWSs) are designed to identify patients at risk of deterioration. The Modified Early Warning Scores (MEWS), for example, utilizes patient data including blood pressure, heart rate, respiration rate, body temperature and patient’s responses in order to identify patients according to their severity. Clinical staff can use the MEWS score as a guide to monitor the patients and give help immediately if needed. This research focuses on the development of machine learning algorithms for the identification of deteriorating patients in intensive care wards with physiological parameters acquired from the hospital information system and the central patient monitoring system. In contrast to other studies in which the algorithms were developed only on public datasets, our study utilizes both public datasets and local clinical dataset from Songklanagarind’s Hospital with an aim to make our algorithms work in local narratives. In the first part, we proposed a data-driven framework for predicting the risk of mortality that combines high-accuracy recurrent neural networks with interpretable explanations. Our model processes time-series of vital signs and laboratory observations to predict the probability of a patient’s mortality in the intensive care unit (ICU). We investigated our approach on three public critical care databases: Multiparameter Intelligent Monitoring in Intensive Care III (MIMIC-III), MIMIC-IV, and eICU. Our models achieved an area under the receiver operating characteristic curve (AUC) of 0.87–0.91. Our approach was not only able to provide the predicted mortality risk but also to recognize and explain the historical contributions of the associated factors to the prediction. The explanations provided by our model were consistent with the literature. Patients may benefit from early intervention if their clinical observations in the ICU are continuously monitored in real time. In the second part, we used the recurrent neural network (RNN) model to predict the risk of occurrence of clinical conditions. Our results varied with an area under the receiver operating characteristic curve (AUROC) of 0.55-0.73 depending on the clinical condition. Our model can provide interpretive data to explain the influence of model prediction parameters and increase confidence in the model. The clinical staff was informed before the clinical conditions and could determine the treatment plan to achieve a better outcome for the patient.