Applying Machine Learning to Heart Disease Risk Analysis

Abstract

Heart disease is known to be deadly; about 700,000 Americans die from it annually. It is essential for young adults to calculate their risk early on, as many factors could lead to a high probability of heart disease. Some factors, such as diet or weight are obvious, but several others, such as mental health, hours of sleep, and even race, can be crucial indicators. This is where machine learning can help, as AI models can look at a large number of factors together to calculate risk by drawing upon real-world data. This paper discusses the implementation of one such model. The model described below is trained on data from about 320,000 patients in the United States. The model uses a Sequential Neural Network algorithm and was created using TensorFlow. It analyzes each patient's data and calculates their risk for heart disease, performing reasonably well with a 77% accuracy. This model can replace a linear-regression-based model, such as the ASCVD Risk Estimator, as it accounts for more factors in its analysis. Further testing with new and diverse datasets and different algorithms, like a RandomForestClassifier, could potentially improve this model.