Machine learning based feature optimization and early detection system in heart diseases

Abstract

In today's world, humanity faces a myriad of challenges, many of which pose significant threats to our well-being. Chief among these challenges are health-related issues. Among these health problems, heart diseases stand out as the leading cause of mortality. Consequently, the early diagnosis of heart diseases plays a pivotal role in mitigating mortality rates and enhancing people's overall quality of life. This study aims to employ machine learning algorithms to enhance the early detection capabilities of heart disease. A dataset comprising the health records of 253,680 patients with heart disease is analyzed using five distinct machine learning algorithms: Logistic Regression, K-Nearest Neighbors Classifier, Decision Tree Classifier, Naïve Bayes, and Linear Support Vector Machine (Linear SVM). The dataset is partitioned, with 80% allocated for training the algorithms and the remaining 20% for testing. Furthermore, the study's evaluation employs four different metrics: accuracy, precision, recall, and the F1measure. Initially, early diagnosis of heart disease is attempted using the complete set of features in the dataset. However, this approach results in excessive costs and time consumption. Subsequently, a feature reduction process is implemented to optimize resource utilization, yielding an improved early detection rate. The research findings indicate that Logistic Regression outperforms the other algorithms, achieving the highest success rate with an accuracy score of 90.67%. These research results underscore the substantial contribution of machine learning algorithms to the early detection of heart disease, ultimately enhancing the quality of life for individuals.