In the fourth stage, the best performing network obtained in the second stage was retrained on all data to increase the prediction accuracy and
evaluate how well the ANN model characterized the desired behavior. Retraining the network with all datasets is expected to provide
reliable predictions and better accuracy measures. It has been presented through several research studies by Najjar et al.8, 9, 10 that stage
four is recommended to arrive at a better performing network model. The optimal network structures for the static ANN models were
selected based on statistical measures such as mean root square error (MRSE), mean absolute relative error (MARE), and coefficient
of determination (R2). The final architectures of the developed auto-associative networks are listed in Table