Classification of SAECG by autoregressive modelling and neural networks

The paper describes investigations into the classification of signal-averaged electrocardiogram (SAECG) signals, with regard to detection of the onset of hypoglycaemia in diabetic patients. Firstly, feature extraction is carried out to obtain time-domain features, which are classified by neural networks. Secondly, the SAECG signals are modelled by autoregressive modelling (AR), and the parameters classified using linear discriminant analysis. The classification performances using both approaches are compared. ECG datasets were obtained from ongoing related research, and consist of paired ECG-glucose readings from type-1 diabetic patients. Data was recorded overnight in the patient's own homes.