What is the clinical evidence from this study?

Study design: Other. Population: Arrhythmia (n=47). Intervention: Random Forest with Fast Fourier Transform feature extraction vs. Decision Tree, Logistic Regression, SVM, KNN. Primary outcome: Classification accuracy.

January 1, 2024Open Access

Comparative Analysis of Machine Learning Algorithms With Advanced Feature Extraction for ECG Signal Classification

TSTanuja SubbaSikkim Manipal University TCTejbanta Singh ChingthamSikkim Manipal University

Key Result

A Random Forest machine learning algorithm with Fast Fourier Transform feature extraction achieved an overall accuracy of 98% in classifying ECG signals into normal and pathological classes.

Structured PICO

Population

47 subjects from the MIT-BIH arrhythmia dataset providing 109,446 labeled ECG instances for classification.

Intervention

Machine Learning algorithms (Decision Tree, Logistic Regression, Random Forest, K Nearest Neighbor, Support Vector Machine) with advanced feature extraction (wavelet transform, statistical and frequency-based features).

Comparator

Comparison among the different Machine Learning algorithms.

Outcome

Classification accuracy of ECG signals into 5 classes (Normal, Superventricular ectopic beats, Ventricular ectopic beat, Fusion beats, and Unknown beats).

A Random Forest machine learning algorithm with advanced feature extraction achieved 98% accuracy in classifying ECG signals into five arrhythmia categories.

Limitations

Dealing with missing and highly imbalanced data
Navigating ECG signal complexity and computational complexity

Abstract

Electrocardiogram is a heartbeat signal that can be used for the application of Human-computer interaction. Electrocardiography (ECG) is a prominent way to analyze heart rate and to diagnose cardiovascular disease. However, its availability has been restricted, especially in contexts with limited resources, due to the cost associated with conventional ECG signal processing equipment. The importance of ECG signal processing classification for improving early diagnoses in clinical and remote monitoring contexts is highlighted here. The dataset considered for this work is MIT-BIH arrhythmia which has 15 categories and summarized in 5 classes Normal (N), Superventricular ectopic beats (SVEB), Ventricular ectopic beat (VEB), Fusion beats (F), and Unknown beats (Q). The work discusses the importance of automated classification techniques that make it possible to analyze vast amounts of ECG data effectively and objectively. This research presents an investigation into the classification of ECG signals using various Machine Learning (ML) methods. Specifically, the performance of Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), K Nearest Neighbor (KNN), and Support Vector Machine (SVM) algorithms are examined. Among these classifiers, RF exhibits a remarkable accuracy of 98%. The results demonstrate the superior performance of the proposed approach for heartbeat classification systems.

Demander à l'IA

Bookmark

View Full Paper