Empirical Analysis of Machine Learning Algorithms on Imbalance Electrocardiogram Based Arrhythmia Dataset for Heart Disease Detection

• Published in: Arabian journal for science and engineering (2011) Vol. 47; no. 2; pp. 1447 - 1469
• Main Authors: Ketu, Shwet, Mishra, Pramod Kumar
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2022; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Arrhythmia; Balancing; Cardiac arrhythmia; Cardiovascular disease; Datasets; Decision trees; Developing countries; Diagnosis; Disease prevention; Electrocardiography; Empirical analysis; Engineering; Hazards; Health services; Heart diseases; Humanities and Social Sciences; LDCs; Machine learning; multidisciplinary; Research Article-Computer Engineering and Computer Science; Science; Support vector machines; Heart disease detection; Electrocardiogram (ECG); Class imbalance; Empirical analysis; Machine learning
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-021-05972-2

Living beings are subjected to many hazards during their course of life. Owing to high mortality rate, heart disease (HD) is among leading hazards for living being. It is world’s one of the critical disease due to its complex diagnosis and expansive treatment. It has predominantly affected the health care sector of developing as well as developed countries. Inadequate preventive measures, diagnosis shortcomings, inefficient medical support, lack of medical staff and advancements have led to severe impacts on developing countries. The paper exhibits state-of-the-art of various intelligent solutions for HD detection with an empirical analysis of machine learning algorithms on electrocardiogram-based arrhythmia dataset for disease detection. A critical investigation is being performed using eight machine learning algorithms, Support Vector Machine, K-Nearest Neighbors, Random Forest, Extra Tree, Bagging, Decision Tree, Linear Regression, and Adaptive Boosting, under imbalanced and balanced class paradigms. The performance of these algorithms is tested with four metrics namely, precision, recall, accuracy, and f1-score. The empirical analysis presents an interesting insight on the structure of dataset. Initially for binary class balancing problem majority class have more accuracy than the minority class because model’s training dataset is crowded with majority class tuples than minority class. The paper uses Synthetic Minority Over-sampling Technique for data balancing. It has not only increased the overall accuracy of the algorithm but also the individual accuracy of the classes. Hence, the accuracy of the minority class will not be sacrificed.