Dear Tadele,

It is a good idea to create a hybrid method of feature selection that combines ANN with Genetic Algorithm (GA). It is known that ANN solutions are the local minima, to avoid this issue you can start a supervised ANN with an initial solution which can be the global optimal solution provided by GA . You can also use the solution provided by unsupervised ANN such as SOMs to create the population of GA in order to find the global optimal solution. 

Check the following publications for example about using ANN and GA.

https://www.researchgate.net/publication/3449931_Multicomponent_Image_Segmentation_Using_a_Genetic_Algorithm_and_Artificial_Neural_Network

Best regards

Article Multicomponent Image Segmentation Using a Genetic Algorithm ...

Dear Deressa,

                      I have a suggestion which is to take the sonograph of the fetal heartbeat over the entire curve of build up of and subdisizing of amniotic fluid. I think you will be able to CTG plot the development of the fetus which will be the plot of the Genetic Algorithm using the ANN version of the foetal development String (DBranes String Mallick-Hamburger-Mallick Functor Algebra Calculus) (Mallick, Hmaburger, Mallick (2016, 2017, 2017)).

Soumitra K Mallick

The hybrid method of ANN and GA posses significant impact on the bio process optimization studies. Methods   suchs as PBD and RSM can nor adress the nonliner relation ship between casual factors and response variables. ANN stimulates the process in a way of Central Nervous System (CNS) function and predict the non-linear relationship between casual factors and response variables. Hybrid ANN-GA has a remarkable ability to derive meaningful information from complicated data and predict the genetic fitness of the studied organism in the fermentation process

Dear Awad, Soumitra,Soni,Upendra,

I want to  appreciate for your cooperate description and immediate answer. I was so glad specially the reference you provided and informed me to read . Please,I request your continual contribution on doubt i will forward in my research . Lastly ,this is my great pleasure being you are volunter problem solver in my subject area...

               For your appreciation ......

                   Best Regards

Tadele Deressa

You seem to be under the very widespread misconception that the purpose of training a neural network is to find the absolute minimum of the cost function. The purpose of training a neural network is to find a model that is able to generalize successfully to data that is not present in the training set. In the vast majority of cases, the model whose training error is minimum is overfitted to the data, hence generalizes poorly, while a model whose training error is a local minimum gives excellent results.

My advice is:

1) Do a thorough literature serach. There are zillions of papers that are based on the above misconception, and many of them used the CTG dataset. There is no point in replicating research that has been carried out before.

2) Forget about genetic algorithms or any other method (e.g. simulated annealing) that tries to find the absolute minimum of the training cost function. Use a good, fast second-order optimization method (e.g. Levenberg Marquardt, BFGS), make many trainings with different initializations of the parameters, and use a good, statistically meaningful, model selection method to find the model that generalizes best given the data that is available to you. Use a good, statistically meaningful, feature selection method.  For a full neural network design method see: Neural networks: methodology and applications (Springer 2005)

You can incorporate GA in several aspects of the training of ANN: feature selection, determination of synaptic weights, and initialisation of some parameters (for example, cluster centres and widths if the ANN model that you use is RBFN or WNN). From my reading, the hybridisation of any metaheuristic algorithm (GA, PSO, ABC, etc) with ANN will work well.

Dear Gerard , Kee

Thank you  for your contribution and addressing important information...