Dear Hariharan Subramanian,

I suggest you to see links on subject.

https://www.diva-portal.org/smash/get/diva2:806491/FULLTEXT01.pdf

https://medium.com/@duoduoyunnini/introduction-implementation-and-comparison-of-four-randomized-optimization-algorithms-fc4d96f9feea

https://www.optibus.com/solving-np-hard-problems-to-optimize-large-scale-systems-it-sounds-complicated-but-its-not/

https://cs.stackexchange.com/questions/67938/decisional-problems-vs-optimization-problems-np-complete-vs-np-hard

Best regards

This can be useful:

https://scholar.google.com/citations?view_op=view_citation&hl=en&user=z_1xf2wAAAAJ&citation_for_view=z_1xf2wAAAAJ:2osOgNQ5qMEC

I think the simplest answer is that there is no such "best" algorithm which gives better results than all other algorithms. Most often, we are satisfied by a good answer and heuristics often give good solutions that are near-optimal.

C.P. Ravikumar

You are right, but say I am working on a scheduling problem, and they have used GA on majority of papers.

Now say I have read some papers on "Comparison of EAs" and I have found that algo X is winning in majority of cases.

Is it a good idea to implement X to the scheduling problem ? Just because it has won on some benchmark functions and problems? Or How do you evaluate whether an algo is designed/suited for problem

I have heard from some profs that the above is not a good strategy as :

1. Just because an algo wins over others inn a comparison paper, doesn't mean its a universal winner.

2. There are certain algorithms that are designed for your problem. Those shall usually win over any other heuristic.

My question is:

How to succeed in point 2? I am currently working on airline crew pairing.

the literature has applied the following:

1.Branch & cut, branch and bound, branch & price, column generation

2. Genetic algo

3. Simulated annealing

4. Tabu search

Majority have conducted extensive research on column generation.

Now, A prof has said that it is well defined/concluded in the literature that standard IP algos have won many times over GAs/other metaheuristics.

But I still feel there might be a goldmine.

In a recent comparison paper, it showed that CS(Cuckoo search)and DE(differential evolution) were winning majority. I know that applying an algo X to problem Y is not as such a real contribution ..

But How do I assess whether the algo is even designed for my problem? This is the main hurdle I am facing since I am a beginner.

There seem to be many algos winning in their own respective fields, but I am unable to assess if it suitable to even be applied to my problem

Some tricks/key-points with regard to "Identification" shall be really helpful.

Thanks