Hello,
I am working on hybrid neural network. I have combined ANN with two different evolutionnary algorithms to build a prediction model of 6 inputs and one output with one hidden layer of 7 nodes. the problem i am having is the network error or fitness function cannot be minimized. So how can i improve it ?
Thank you for your help.
Best regards
Myriam Letif ,
Can you output the error curve? Also, how are the two algorithms interacting with the ANN?
I think you want to find the optimal weights of your network with 2 evolutionary algorithms. If so, you should know that some of the EA cannot find the optimal solution or converge to it based on the case study (local minima trap). To solve it, maybe it is better to use other EAs or preprocess your original inputs.
If the network can't represent the association, then it's not surprising that the error can't be reduced. Given a network architecture, there exist associations that the network can't represent. This means that there don't exist values for the weights that can make the error vanish.
To represent the association in question it's, therefore, necessary to change its topology, in particular to add additional layers.
- Badges
- Science topic