Prediction of Chaotic Time Series and Resolution of Embedding Dynamics with the ATNN *

Chaotic time series prediction is an essential task of signal processing, physiological analysis and economic forecasting. In this paper, we apply the Adaptive Time-Delay Neural Network (ATNN) to study the chaotic Mackey-Glass differential equation prediction problem. We discuss the embedding dynamics, training set size and performance, and prediction over time. We demonstrate this neural network approach to be flexible, to predict the chaotic dynamics, and to achieve good performance.