Publication

Abstract : In-hospital mortality prediction is a crucial part of healthcare that can significantly impact patient care strategies and clinical decision-making. Employing Machine Learning (ML) methods in the healthcare domain to improve clinical decision-making and create more effective patient care plans, potentially saves lives. One example is the development of a trustworthy model for forecasting in-hospital mortality. In this work, we make use of a diverse set of patient attributes (features), such as demographics, medical history, and physiological metrics, in order to reveal hidden patterns in the patient data. Our goal is to suggest the best machine learning model that can effectively predict in-hospital mortality, eventually improving patient care outcomes. After investigating 7 ML models, namely; Decision Tree (DT), Random Forest Classifier (RF), Linear Regression (LR), Logistic Regression (LogR), Support Vector Machine (SVM), k-NN Classifier and Naïve-Bayes Classifier (NB Classifier), we have compared and concluded that the RF model outperforms the other models, providing us with the best accuracy of 92.662%. In the health care domain, it is very much required to explain why and how a particular ML model predicts the specific outcome, to the healthcare professionals. Hence in this work, Explainable AI (XAI) is used to better interpret the RF Classifier, by generating SHAP (SHapley Additive exPlanations) plots.