Hyperparameter Optimization of AutoRegressive Integrated Moving Average (ARIMA) Model-based Synthesis of Electrocardiogram

Electrocardiogram (ECG) modeling is important for synthesis of the signal for pathology and compressed representation of the signal. Autoregressive integrated moving average (ARIMA) model provides model parameters based on current and previous samples. In this paper, we present the optimization of the ARIMA model hyperparameters using grid search (GS) and particle swarm optimization (PSO) techniques. At first, the preprocessing of the ECG is done using a discrete wavelet transform base approach. Then, the R-peaks are detected along with the cardiac beat extraction. Each cardiac cycle is then modeled using ARIMA model, whose hyperparameters are optimized by GS and PSO. For validation and testing, the different cardiac rhythms (normal and abnormal) and healthy and cardiac diseases are used from mitdb and ptbdb data under PhysioNet. Over six types of cardiac rhythms from the mitdb database, the mean absolute error of
$1.56\times 10^{-2}\ \text{mV}$
and
$0.46\times 10^{-2}\ \text{mV}$
and three types of cardiac rhythms from ptbdb database, the mean absolute error of
$0.29\times 10^{-2}\text{mV}$
and
$0.19\times 10^{-2}\text{mV}$
were obtained with GS and PSO respectively. In general, it was found that the error figures are better for PSO based technique.