Identifying Prognostic Features for Predicting Heart Failure by Using Machine Learning Algorithm
Md. Al Mehedi Hasan, Jungpil Shin, Utsha Das, Azmain Yakin Srizon
Abstract
Being one of the most common cardiovascular diseases, heart failure caused 40 million deaths worldwide in 2015. Previously, various studies have been conducted to predict heart failure at an early stage. Although each study has contributed and continued the development process, a significant breakthrough is still to be achieved. In this research, we focused on finding the features which can predict the death probability at an early stage due to heart failure. Firstly, the dataset was acquired and preprocessed. After that, two feature selection approaches, minimum redundancy maximum relevance, and recursive feature elimination based on Naïve Bayes were employed. Then, five machine learning classifiers, support vector machine, logistic regression, decision tree, naïve bayes and k-nearest neighbors, were utilized. Finally, the performance was measured in terms of accuracy, sensitivity, specificity, f1-score, MCC value and AUC value. It turned out two features which were selected by both feature selection techniques, achieved the highest overall accuracy of 80% for the decision tree classifier. Comparison with the previous best result of 58.5% proved that our proposed methodology can comment with much more certainty that Ejection Fraction and Serum Creatinine are indeed the two factors by which heart failure can be predicted.