Development of an Explainable Prediction Model of Heart Failure Survival by Using Ensemble Trees

Abstract

Cardiovascular diseases (CVD) are the leading cause of death globally. Heart failure prediction, one of the CVD manifestations, has become a priority for doctors, however, up to date clinical practice usually has failed to reach high accuracy in such tasks. Machine learning offers advantages not only for clinical prediction but also for feature ranking improving the interpretation of the outputs by clinical professionals. Thus, the concept of eXplainable Artificial Intelligence (XAI) is aimed to cope with the lack of explainability of machine learning models in the healthcare domain, in this case, and provide healthcare professionals with patient-tailored decision-making tools that improve treatments and diagnostics. This paper presents a heart failure survival prediction model development by using ensemble trees machine learning techniques. Extreme Gradient Boosting (XGBoost) is demonstrated as the classifier with most accurate results (83% accuracy with unseen data) over the other ensemble trees options. Moreover, a features selection preprocessing is made in order to assess which relevant features contribute to the model's results. Next, in terms of improving the explainability of the model developed, a study of features importance is carried out showing the "follow up time period" feature as the most relevant. Finally, a quantitative evaluation of the interpretability and fidelity of the model developed is performed obtaining a balanced ratio between these two indicators.