Abstract: Objective To construct a model of heart failure risk prediction based on four machine learning algorithms in order to support early diagnosis and intervention. Methods After reviewing the heart failure dataset published on the Kaggle community, feature selection was used to select relevant factors related to heart failure as predictive indicators. Four machine learning algorithms, namely logistic regression, support vector machine, random forest, and XGBoost were selected to establish predictive models. Compared and analyzed its accuracy, precision, recall, F1 score and area under the ROC curve (AUC) to verify the performance of the model. Results The study analyzed 11 features of 918 patients with heart failure and selected 10 feature factors for modeling. After optimizing the hyper-parameters through grid search, the XGBoost model performed the best, with accuracy, precision, recall, and f1_score and AUC values were 87.5%, 90.38%, 89.71%, 90.04% and 0.93, respectively. In addition, data analysis showed that exercise ST slope, chest pain type, and exercise induced angina were main influencing factors for heart failure. Conclusions The XG Boost model has the best predictive tool for heart failure, and machine learning algorithms may support early prevention, early diagnosis as well as control of heart failure.

Key words: heart failure, machine learning, prediction