Experiments with a Hybrid-Complex Neural Networks for Long Term Prediction of Electrocardiograms

In this paper we present the results obtained by a partially recurrent neural network, called the hybrid-complex neural network (HCNN), for long-term prediction of electrocardiograms. Two different topologies of the HCNN are reported here. Even though the predicted series were not similar enough to the expected values, the HCNN produced chaotic time series with positive Lyapunov exponents, and it was able to oscillate and to keep stable for a period at least 3 times the training series. This behavior, not found with other predictors, shows that the HCNN is acting as a dynamical system able to generate chaotic behavior, which opens for further research in this kind of topologies.