Hello,

For regular clustering, you can pretty much always forget about the evolutionary approaches. Any other type of clustering, distance based, density based, distribution based, model based or whatever fits you data best will be a more solid approach.

Not that you cannot use an evolutionary approach for regular clustering, it works. But why would you want to adapt an evolutionary approach that give approximate solutions when you can use an already existing algorithm that does the same thing just fine and has been proved to converge ?

Keep the evolutionary approaches for cases where there are too many solutions to explore and there is no exact algorithm doing what you want in a reasonable time. For instance, evolutionary approaches are fine for ensemble learning clustering.

PS : You can have a distribution based evolutionary approach. For instance if you use genetic algorithms, at some point you have to define a fitness function, and you can define it based on data distribution.

For Type-2 fuzzy clustering you can use an evolutionary algorithm to find the best secondary membership function, but it makes system slow.

Also, I suggest you to study about  imperialist competitive algorithm. It is an evolutionary approach which use in several clusterng algorithms. e.g.:

http://link.springer.com/article/10.1007%2Fs12046-011-0026-4

http://www.academia.edu/490680/An_efficient_hybrid_algorithm_based_on_modified_imperialist_competitive_algorithm_and_K-means_for_data_clustering

https://www.academia.edu/490688/A_New_Hybrid_Algorithm_Based_on_Imperialist_Competitive_Algorithm_and_k-means_for_Cluster_Analysis

https://www.academia.edu/490666/Log-Spectrum_based_RSTB_Invariant_template_matching_with_Modifed_ICA

I mostly agree with Jeremie (no real benefit of evolutionary approaches) but I wonder which provable convergence (of alternative methods) to an exact (?) solution he is referring to. Considering k-means as a prototypical example, this method only returns a locally optimal solution (although often satisfactory in practice).

The mathematical objective k-means and related distance based methods are trying to optimize is to find a partition of a set into k subsets such that the sum of distances of each point to its respective mean is minimal. This optimal partitioning problem is NP.  

fixed. A proof of convergence is not a proof of convergence toward an exact solution.

Try AHFCM method as an automatic clustering method uses the histogram based vision to clustering data.

Thanks Jeremie, I was thinking the same. I recently used Multi-Objective Evolutionary approach to develop Ensemble based clustering. I also have found out by experiments that Evolutionary approaches to be useful in clustering ensembles. 

The validation results of two clustering algorithms are available in the following publication: 

Article Automatic building extraction in dense urban areas through G...