Nonlinear Prediction for ECG by 2^<nd>-order Volterra Neuron Network

Abstract

The prediction accuracy of QRS complex is not good in predictions of electrocardiogram used Volterra functional series yet. The cause is not to be able to predict enough, even if Volterra functional series is used because production mechanism for QRS complex contains strong nonlinearity. Therefore, means for prediction using an input delay neuron network which nonlinear prediction capability is higher was proposed. However, it is not sufficient on QRS complex though the prediction accuracy is improved overall. Aim of this study is further to improve on prediction accuracy for electrocardiogram using a 2^<nd>-order Volterra neuron network which processing performance to time series signal with strong nonlinearity is high. Then, nonlinear prediction capabilities for electrocardiogram by an input delay neuron network and a 2^<nd>-order Volterra neuron network are evaluated. As a result, prediction accuracy of the 2^<nd>-order Volterra neuron network is shown the remarkable improvement of 66.7% than the input delay neuron network on the average about root mean square error.