We demonstrate the performance of the method by experimenting with the chaotic Mackey–Glass univariate time series dataset which is produced by nonlinear time delay differential equation. To have fair comparisons with earlier works, we define the output of x(t+6)with the input variables of x(t), x(t−6), x(t−12)and x(t−18). We considered the RMSE as the performance index (as em-ployed in other works [10,11]) for this training and testing process of this dataset. The compared result is shown in Table1and Fig.3.
The present results obtained by the proposed model which has less complexity with one hidden layer and three nodes in the hidden layer compared to the others confirms the capability of the method. The column chart in Fig.3compares the training and testing errors of the methods sorted by reducing testing error. Our method (number 17) outperforms 10 methods in the literature out of 17 and is comparable with the rest of the models results while the present model is less time consuming for it is less com-plex.