Abstract

Cardiovascular diseases have become one of the major sources of morbidity in the world, therefore early detection of the people at risk would allow medical professionals more time to act. One of the approaches to the problem through ML is the use of the information that includes such variables as age, blood pressure, total cholesterol, blood glucose, heart rate, symptoms of the patient, and test results. In this paper, we give a brief description of the most popular algorithms for predicting the risk of heart diseases using ML: Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, Naïve Bayes, K-Nearest Neighbors, and Neural Networks. Deep learning, Explainable AI (XAI), small data, class imbalance, bad data quality, privacy, and hospital validation will also be discussed. This narrative literature review is based on the synopsis of the project, some pieces of literature, and dataset documentation. There was no training of any ML model nor conducting of any experiments. As it follows from the papers under study, ML allows to separate patients into risky and non-risky categories; however, the results strongly depend on the selected dataset, goal of prediction, validation method, and patients themselves. Results from small benchmark datasets tend to be highly optimistic and cannot be used to predict clinical relevance of the models. We also present a process flow of how the data should be prepared, modeled, evaluated, interpreted and externally validated and hope that such models will help physicians more than replace them.

Keywords

How to Cite This Article

Rights