Application of Machine Learning for Heart Failure Prediction

Cardiovascular disease is one of the more lethal diseases worldwide, and its main manifestations are myocardial infarction and heart rate failure. Fortunately, most cardiovascular diseases can be prevented to some extent through their risk factor analysis. In this paper, we propose a new feature selection scheme by integrating cox proportional hazards model and random forest models to explore the existing data experimentally. This paper integrates two types of models by a multiphase data fusion strategy in order to mitigate selection difficulties from different feature rankings, particularly in small circumstances. We also provide visualization techniques to improve the transparency, trustability, and interpretability. This scheme, combined with neural networks to construct a survival analysis model of heart failure, has two crucial achievements. First, models result statistical significance on features. Second models can help assess the impact of different factors on heart rate failure and predict future risk profiles, providing powerful new support for the prevention and treatment of heart failure. In the study of 299 patients, the feature selection scheme single out age, ejection fraction, and serum creatinine as relevant features, through which the model achieve a relatively high predictive accuracy with C=0.7544 (95% CI: 0.746S-0.7620). With a forementioned performance, this model vitally aids to clinicians with undetermined patient data characteristics. Furthermore, the model improves accuracy by about seventeen percent comparing to the human benchmark C = 0.59 (95% CI: 0.53-0.65), by about eighteen percent comparing to the two-dimensional model, and also has a better generalization ability comparing to the Multidimensional machine model.