Neural network is just a regression based program. The accuracy improves with number of training data sets. Even if your network is robust with large data sets used for training and testing, it is very unlikely that the results match with the numerical analysis outcome. A numerical model has a set of assumptions and the output variables follow a set of scientific equations derived from fundamentals, whereas a ANN model has no science behind and it only works based on pattern recognition from previously available knowledge. Hence they both can differ significantly.
Dear Bhavnesh Kumar, One reason may be that the training process of your deployed neural network was not rigorous enough. In other words, for a neural network's performance to be consistently high, it must be thoroughly trained with the right set of parameters. So you may need to consider this point in your project. I hope you would find this helpful.
Neural network is just a regression based program. The accuracy improves with number of training data sets. Even if your network is robust with large data sets used for training and testing, it is very unlikely that the results match with the numerical analysis outcome. A numerical model has a set of assumptions and the output variables follow a set of scientific equations derived from fundamentals, whereas a ANN model has no science behind and it only works based on pattern recognition from previously available knowledge. Hence they both can differ significantly.
I agree with Profs. Madhavi and Oluwarotimi, you should go to the begining and redesign your ANN, select an appropriate data set and training procedure.