Machine Learning Techniques for Heart Failure Prediction

Abstract

This paper discusses the performance of four popular machine learning techniques for predicting heart failure using a publicly available dataset from kaggle.com, which are Random Forest (RF), Support Vector Machine (SVM), Naive Bayes (NB), and Logistic Regression (LR).  They were selected due to their good performance in medical-related applications.  Heart failure is a common public health problem, and there is a need to improve the management of heart failure cases to increase the survival rate.  The vast amount of medical data related to heart failure and the availability of powerful computing devices allow researchers to conduct more experiments. The performance of the machine learning techniques was measured by accuracy, precision, recall, f1-score, sensitivity, and specificity in predicting heart failure with 13 symptoms or features. Experimental analysis showed that RF produces the highest performance score, which is 0.88 compared to SVM, NB, and LR.  Further experiments with RF were also conducted to determine the important features in predicting heart failure, and the results indicated that all 13 symptoms or features are important.