Applying Propensity Score and Support Vector Machine to Construct a Predictive Model for Heart Disease

Abstract

Exercise ECG is currently the best way for diagnosing heart disease, but it is not suitable for everyone. This study used data mining to establish a model to predict the risk of heart disease. Maximum oxygen uptake (VO2max) was used as an indicator of determine that the person was a high-risk or low-risk heart disease patient. Data of the National Health and Nutrition Examination Survey from the United States were used in this study. Due to scattered distribution of the data, which diminished the prediction performance, this study proposed a novel method to stratify data with the propensity scores. The subsets of data were trained by the support vector machine to establish the prediction model. The results of this study showed that the model had an AUC of 0.899. Our model can make a more accurate prediction to identify whether a patient has a higher risk in heart disease.