If we were in the nineties or at the beginning of this century, I'd centaintly said yes. But the use of metaheuristics and other optimization techniques to optimize ANN, both the architecture and the weight, is a problema widely studied. In that situation, I think currently it is difficult to make a good contribution to be published in a prestigious/international journal, reaching the level to defend a PhD dissertation based on that contribution.

There is of course lots of work and PhDs still being done in this area, but it is a heavily studied area. There would have to be something particularly novel about the application, analysis, or results. Conversely, it's not necessarily a bad place to start a PhD since preliminary results and progress should be quite easy to attain, and there is lots of relevant literature to follow.

Syed,

The good thing that you have with a topic like this is that you can do a good justification of the problem (it has been studied extensively), you can do an extensive review for your contribution and you have solid starting benchmarks. The hard part is as Sebastian and Stephen pointed out is to make a novel contribution. Also obtaining better results than current state of the art is difficult but not an impossible problem.

Thanks a lot, I have started working on BPNN hybridization and recently, I got a whole lot of comments...But the problem was that those comments made me confused further, instead of being helpful.

You will need to determine whether you are using an existing philosophy and possibly gain an incremental improvement on an existing problem. This might warrant a Ph.D. However, if you can find a new paradigm or philosopophy for NN optimization rather just an incremental benefit in performance, this should lead to a larger impact. As said above, the area is well-studied and may not lend itself to such paradigm shift easily.

Thank you Kris :D

I want to confirm the opinions that were mentioned so far about this area is well-studied. Nevertheless, there are area that was not well studied enough. Therefore, I want to give you my advice: please, focus your attention on the problem "to find quality lower bound LB(D) of objective function", LB(D)

Yes, finding the quality bound of an objective function is a significant contribution. The bound must be expressed using measurable properties of the problem. To do so, You need to model the problem ( i.e using discrete mathematics) including the quality bound of the objective function and examine the relationships among the bound and the various properties.

An interesting answer to your question is provided by Sörensen, K., 2013. Metaheuristics - the metaphor exposed, International Transactions in Operations Research, 1, 1-16. DOI: 10.1111/itor.12001