Identification and Validation of Prominent Features for Predicting Mortality in Heart Patients with Left Ventricular Dysfunction Using Machine Learning

Abstract

AbstractMachine Learning (ML) is a strong tool for medical prognosis, and it has the potential to give this branch of medicine a huge boost by allowing doctors to make accurate predictions about a patient’s future health using various forms of medical data. ML algorithms have proven to be reliable and effective in decision making with good classification accuracy. They can model nonlinear relationships, which are frequent in medical data, and apply them to predictive tasks such as forecasting a future event. In this paper, an attempt has been made to predict the mortality of heart patients with left ventricular dysfunction. Feature selection methods have been used to rank the input features in the dataset and identify four prominent features. Different combinations of these prominent features have been applied to five ML algorithms namely, Decision Tree, Gradient Boost, Random Forest, Support Vector Machine and k Nearest Neighbors to find the best performing combinations using F1-Score and AUC ROC. Considering additional performance parameters, further analysis is carried out to identify the best feature combination and the most effective ML algorithm for predicting mortality and the results are provided for the same.KeywordsCardiovascular disorderClassificationDecision treesMachine learningMedical prognosis