A Novel Method for ECG Classification Using Polynomial Based Curve Fitting

The recent biomedical signal processing techniques are becoming the main part of real time electrocardiography (ECG) signal analysis. The classification of arrhythmic condition is the main process carried out in ECG signal processing. This process allows us to evaluate the abnormalities that occur during disturbances in heart rhythm. In this paper, the preprocessed ECG signals are classified using curve fitting and random forest-based algorithms into normal and abnormal signals. Initially the noise from the signal is removed using FFT technique, then for the detection of R-peaks windowing technique and thresholding technique are used. A novel polynomial-based curve fitting method is applied to the extracted HRV signal for obtaining the features for classification. The computed features are both statistical and wavelet features. They are utilized in various combinations for further classification of the input ECG signals using different algorithms. The arrhythmia classification system proposed here is applied to input ECG signals obtained from MIT-BIH Arrhythmia database, as well as several other international ECG signal databases. The highest performance shown by random forest algorithm in this system, makes it more compatible to work very fast and classify with greater accuracy. The random forest algorithm can be useful for long term ECG beat classification and disease diagnosis. The experimental results bring an overall improved accuracy, sensitivity and specificity with a highest percentage of 98.82% in comparison to the other approaches that are existing in machine learning techniques.