A machine learning approach using DWT, ICA, and an SVM quadratic kernel classified five classes of ECG arrhythmia beats with an overall average accuracy of 98.49%.
A machine learning approach using DWT, ICA, and SVM achieved 98.49% overall accuracy in classifying five types of ECG arrhythmia beats from the MIT-BIH database.
Electrocardiogram (ECG) remains the most reliable and low-cost diagnostic tool to evaluate the patients with cardiac arrhythmias. Manual diagnosis of arrhythmia beats is very tedious due to the nonlinear and complex nature of ECG. Likewise, minute variations in time-domain features viz. amplitude, segments and intervals are difficult to interpret by naked eye. The current paper, describes a machine learning-based approach for computer-assisted detection of five classes of ECG arrhythmia beats using Discrete Wavelet Transform (DWT) features. Further, methodology comprises dimensionality reduction using Independent Component Analysis (ICA), ten-fold cross-validation and classification using Support Vector Machine (SVM) kernel functions. Using ANOVA significant features are selected and reliability of accuracy is measured by Cohen's kappa statistic. Large dataset of 110,093 heartbeats from 48 records of MIT-BIH arrhythmia database recommended by ANSI/AAMI EC57:1998, which are grouped into five classes of arrhythmia beats viz. Non-ectopic (N), Supraventricular ectopic (S), Ventricular ectopic (V), Fusion (F) and Unknown (U) are classified with class specific accuracy of 99.57%, 97.91%, 92.18%, 76.54% and 97.22% respectively and an overall average accuracy of 98.49%, using SVM quadratic kernel. The developed methodology is an efficient tool, which has intensive applications in early diagnosis and mass screening of cardiac health.
Desai et al. (Tue,) conducted a other in Cardiac arrhythmias (n=48). Machine learning-based approach using DWT, ICA, and SVM was evaluated on Overall average classification accuracy of five classes of arrhythmia beats. A machine learning approach using DWT, ICA, and an SVM quadratic kernel classified five classes of ECG arrhythmia beats with an overall average accuracy of 98.49%.