Early arrhythmia prediction based on Hurst index and ECG prediction using robust LMS adaptive filter

This paper aims to early arrhythmia prediction and investigate the use of robust adaptive filters to forecast the ECG signal. Different robust adaptive filters are examined for ECG prediction. Features in time and time-frequency domains have been extracted, and the Hurst index has been calculated in two domains. The performance of the SVM, KNN, and the ensemble of LogitBoost trees for model construction has been examined for detecting the occurrence of an arrhythmia in the predicted ECGs in an inter-patient scenario. Results show that pseudo-Huber adaptive filter is the best choice for ECG prediction. Also, classification performance measures besides the McNemar test show that the predicted signal is suitable to use for early arrhythmia detection with accuracy, precision, sensitivity, and specificity of at least 98%.