A Data-Driven Study of Prediction Methods for Coronary Heart Disease

Coronary heart disease (CHD) is a globally recognised, highly prevalent disease with a high risk of death and a low cure rate. The World Health Organization estimates that deaths from heart disease will reach 23 million by 2030. Therefore, it is imperative to find a fast and effective method for early diagnosis in order to provide patients with early intervention and improve the effectiveness of treatment. With the in-depth development of machine learning, the function of data analysis and prediction will efficiently help doctors to make a preliminary cluster for a large number of people and detect those who have a dangerous rate of developing coronary heart disease. In this paper, three data pre-processing methods, Smote, Borderline Smote and K-means Smote, were used to construct a risk prediction model for coronary heart disease (CHD) based on an unbalanced data set, combined with four algorithms, Logistic Regression, Random Forest, KNN and SVM. After analysing the data characteristics and adjusting the parameters, different combinations of these methods were compared and a better classification method was selected to predict CHD, achieving higher accuracy, precision, AUC and f1 score. Overall, through experiments, the random oversampling and SMOTE methods can effectively solve the data imbalance problem in most cases.Our final training accuracy could be up to 99%, and the testing accuracy could reach 93%.