DFS-Enhanced Lightgbm: An Extended Lightgbm Model Applied to ICU Heart Failure Mortality Prediction

Heart failure (HF) is a pervasive clinical syndrome, and there is a demanding innovative approach to mortality pre-diction in Intensive Care Units (ICUs). We utilized data from 1177 adult HF patients from the MIMIC-III database and generated a comprehensive set of new features using DFS technology. Our in-novative approach synergistically combines DFS and LightGBM, enhancing the prediction of heart failure mortality in ICUs by efficiently handling large datasets and capturing intricate feature interactions. Compared to conventional XGBoost and LightGBM models, our approach achieves superior performance, with an accuracy of 0.954173, recall of 0.953177, and an F1 score of 0.953377, alongside a minimal mean square error of 0.0458265. Additionally, we have incorporated medical expertise to elucidate the implications of the top features, illustrating the potential of our approach in providing both high predictive performance and valuable clinical insights.