A Comprehensive Investigation of the Performances of Different Machine Learning Classifiers with SMOTE-ENN Oversampling Technique and Hyperparameter Optimization for Imbalanced Heart Failure Dataset

Abstract

The chronic cardiac condition myocardial infarction (heart failure) is characterized by decreased blood supply to the body as a result of the heart muscles’ impaired contractile properties. Patients with heart failure, like those with any other cardiac disorder, have difficulty performing daily activities and have a shorter life expectancy, with the vast majority of cases resulting in death at some point during the patient’s lifetime. Treatment outcomes and patient quality of life improve significantly when patients with heart failure are identified early and are likely to survive. As a result, machine learning techniques can be extremely beneficial in this situation because they can be used to predict the survival of heart failure patients in advance, allowing patients to receive the most appropriate treat- ment at the earliest possible stage. As a result, six supervised machine learning algorithms were applied to a dataset of 299 people from the University of California, Irvine Machine Learning Repository in order to predict their chances of surviving heart failure. There were a variety of algorithms used in this study including Decision Tree Classifier, Logistic Regression, Gaussian Nave Bayes, Random Forest Classifier, K-Nearest Neighbors, and Support Vector Machine, among others. Prior to scaling the data, a preprocessing step was carried out, and both the standard and min-max scaling methods were employed. When it came to optimizing the hyperparameters, the techniques grid-search cross validation and random search cross validation were combined. Data resampling techniques such as the edited nearest neighbor (SMOTE-ENN) and synthetic minority oversampling (SMOTE) data resampling are also employed (SMOTE-ENN). It has been thoroughly compared and analyzed the outcomes of all of the different approaches. As a result of these findings, the Random Forest Classifier (RFC) outperforms all other approaches, achieving a test accuracy of 90 percent when compared to the other approaches when SMOTE-ENN and the standard scaling technique are employed. With the help of an imbalanced dataset, this comprehensive investigation vividly illustrates the application and compatibility of sev- eral machine learning algorithms. Among the methods for improving the performance of machine learning algorithms discussed in this investigation are the SMOTE-ENN algo- rithm and hyperparameter optimization.