A Comparative Study of Machine Learning Algorithms for Heart Failure Survival Prediction

Abstract

Heart failure (HF) is one of the most common causes of death and morbidity in the world and poses to be a serious problem in early diagnosis and survival prognosis. In this study for predicting heart disease survival using a dataset of 5000 patients. Precise and early prognostication potential in automating and improving survival analysis. This paper documents the comparison of different ML techniques applicable to predict survival among HF patients: Random Forest, Decision Tree, Gradient Boosting, K-Nearest Neighbours, Support Vector Machine, Ad Boost, Logistic Regression, and Naive Bayes. This study will be based on the data that we will use to include some of the clinical parameters that were read by the patients who had heart failure. Before training the models, data pre-processing, balancing with ADASYN and feature scaling have been used. The assessment was done based on standard metrics of performance, including accuracy, precision, recall, F1-score, and ROC AUC. Model performance was analysed using visualization tools such as a confusion matrix, ROC, and importance of features plots. In this study, using analogical algorithms depends on accuracy, precision, recall, F1-score, and Random Forest (RF) shows the highest accuracy of of survival events among patients with HF also continues to be an imminent obstacle because of the heterogeneous and complex characteristics of the disease. Nonetheless, the current developments in machine learning (ML) have demonstrated 99.5%.