Multinomial Logistic Regression Classification Model for Arrhythmia Detection

Abstract

In the present medical era, Electrocardiography has been used to record heart activities for the basic visualization of various cardiac diseases. ECG signal properties analysis for effective arrhythmia classification has been considered a challenging issue due to the interruption of various sorts of noises that includes baseline wander, power line interference, and motion artifact noise. This challenge has been addressed in this research through the Multinomial logistic Regression (MLR) classification model. This mathematical model has been structured for the effective removal of various noises and to improve the classification ratio for effective arrhythmia detection from ECG signals. In MLR, Fractional Wavelet Transform is used for preprocessing of ECG signal for removing the noise and to determine the QRS interval from the ECG signal. From the Pre-processed signal Stacked Autoencoder (SAE) is used to validate the dimensionality of the retrieved features for effective prediction and classification of multiple forms of arrhythmias. Based on the extracted features of the ECG data MLR classification model obtains a maximum classification ratio for accurate arrhythmia identification. The experimental findings show an improved classification ratio of 98.95% with reduced noise factors, Signal to Noise Ratio (SNR) of 35.7dB when compared to conventional algorithms.