Predictive Modeling of In-Hospital Mortality in ICU Heart Failure Patients Using Machine Learning Techniques

Heart Failure (HF) is a complex clinical disease that poses significant risks to public health. According to the European Society of Cardiology, the number of people diagnosed with heart failure has been rising dramatically, with approxi-mately 26 million people affected worldwide in 2024. This trend is expected to increase by more than 50 % by 2030. Given the seriousness of this disease, early prediction of the risk of death in heart failure patients is crucial for developing individualized and effective treatment plans. While many previous studies have primarily focused on building mortality prediction systems with high predictive accuracy, increasing the sensitivity of such a system has not been considered. This study aims to fill that gap by developing a predictive model that achieves high accuracy while also focusing on sensitivity in detecting the risk of in-hospital death for heart failure patients in the Intensive Care Units (ICUs). The model was trained on the MIMIC III dataset of 1177 patients using various Machine Learning (ML) algorithms, with CatBoost being the most effective. This model demonstrated encouraging results, with an accuracy of 90 % and a sensitivity of 81 %. The importance of this sensitivity lies in its ability to determine if a patient requires further intervention before it is too late and to ensure that more high-risk patients are correctly diagnosed; by doing so, healthcare providers can implement timely interventions that could prevent fatalities and improve clinical outcomes, including mortality rates and quality of life for HF patients.