RR interval prediction in ECG signals

Prediction of a signal from recorded time series is always a challenging task. In this paper, the R-R intervals behaviour is estimated using linear and non-linear prediction techniques. The value of each sample point is predicted using a certain number of previous samples and the prediction error is computed. The wavelet transform provides multi-resolution analysis and allows accurate time-frequency localization of different signal properties. This paper presents a nonlinear prediction method from a first order discrete wavelet transform, implemented on artificial neural network based learning structure, compared with an ARMA model based prediction method. The followed parameter is the absolute value of prediction error.