Number of Neurons in a model

Hi everyone,

I have another confusion

Is there any thumb rule or relation between number of neurons in hidden layer and input variables? suppose, I have a data set with 10 input variables. By hit and trial method, how many neurons maximum can I try? I understand that as number of neurons increase, correlation and mse also improves. But at same time, network gets overtrained. How can I be sure that I am not overtraining the network?

Thanks

Arijit

Answer 1

[Ahmad Taher]

Dear Colleague

Hope the following link may help you: http://www.faqs.org/faqs/ai-faq/neural-nets/part3/section-10.html

If you have 10 input variables, I suggest to start with 10 neurons and try different numbers to 20 neurons and then select the best numbers with least MSE.