A Predictive Study of Heart Attack Triggers Based on Machine Learning

Abstract

With increasing health concerns in society and the enormous surge in digital data generated from technological advancements, the need for efficient and accurate methods of value extraction from data is more evident than ever. Thus, such methods are developed to help medical professionals make thoughtful decisions. This paper explores a piece of health data from the Centers for Disease Control and Prevention (CDC) to find the most influential factors contributing to the occurrences of heart attacks and the best algorithm to help predict such occurrences. To achieve that goal, this paper employed algorithms including Support Vector Machine (SVM), Random Forest, and Logistic Regression, along with a technique called permutation feature importance to identify important features. In the end, Random Forest was found to be the most powerful and consistent algorithm to complete the task of prediction in this case, while also showing that average sleep hours, BMI, and mental health status of individuals can be potential indicators of heart attacks. With this result, this paper marks significant features and algorithms that can be subjects of further and more extensive studies, and eventually be applied to the work of real medical professionals.