Performance Evaluation of Classifiers for ECG Signal Analysis

The cardiac well-being of humans can be monitored by non-invasive electrocardiogram (ECG) to a greater extent. Subtle changes in ECG waveform can be identified by computer-assisted tools. Machine learning algorithms play an important role in arrhythmia classification. This paper presents a comparative analysis of various classifiers to support ECG classification. The classification model detects seven arrhythmia types from the generated dataset derived from arrhythmia database of MIT-BIH. The proposed technique considers ECG beat features in time domain based on ECG morphology and statistics. Arrhythmia classification is carried out for seven classes. Performance evaluation is carried out for different classifiers with accuracy, sensitivity, specificity, and F1-score as the evaluation metrics. Classification accuracy up to 97%, Recall up to 92%, F1-score up to 91% and precision up to 91% is achieved with specific classifiers across various arrhythmia classes under consideration.