Mackey-glass chaotic time series prediction using modified rbf neural networks

The characteristics of a nonlinear dynamical system within chaotic system is more intensely studied recently, due to many real-world applications of the nonlinear chaotic system are increasing. For characterizing the ordinary system, usually the relationship between the linearity and the nonlinearity of parameters in the system is needed to be firstly derived, however, creating the mathematical model of the real chaotic system is still a problematic since insufficient basic physical phenomena should be analyzed. Hence, artificial neural networks approach that performed based on nonlinear mathematical model is quite adequate to be used to analyze the chaotic phenomena within the system. Solving the multi-step ahead prediction problem of time series chaotic system is one of the top challenging issues, especially on how to obtain a higher prediction rate. In this paper, a modified Radial Basis Function Neural Network (RBF-NN) is developed and be tested for predicting the future state of a Mackey-Glass equation as the chaotic system. Results experiments show that using training testing paradigm of 50%:50%, the calculated of confidence level accuracy of the neural-predictor system is satisfied for up to 30-steps ahead prediction.