This paper aims to examine the effectiveness of deep learning (DL), a burgeoning aspect of machine learning and artificial intelligence, in exploring input-response dependencies from observed data, especially when complex nonlinearities are present. DL has the potential to be at least as effective as, if not better than, traditional statistical techniques such as response surface methodology (RSM). To test this hypothesis, we developed DL models using Tensor flow and compared their predictions against those of well-established statistical models. Our DL models were hyper parameter tuned using grid search. We found that, for identical input data, DL's predictions closely matched the results of published central composite designs, and often resulted in smaller root mean square errors, indicating greater predictability, particularly in cases where higher order nonlinearities might be present but missed. Therefore, it is recommended that Industrial Engineers, Data Scientists and R&D Professionals incorporate DL in their study of complex processes, along with classical statistical methods, if they have appropriately collected input data. Overall, this study provides evidence that DL can be a valuable tool for exploring complex relationships in data