Following...

Dear Iman, I'm afraid you can't. Garson's algorithm only gives (or at least intends to give) relative importance, not the specific partial derivative results.

Yes, I mean this is so that I am going to obtain the relative importance index for each input of ANN with percent to predict each output. How can I do it?

Just simply substitute the connection weights of your ANN into the formula of Garson's algorithm. The formula can be found in many literatures, i.e. https://www.google.com/url?sa=t&source=web&rct=j&url=http://cs.anu.edu.au/people/Tom.Gedeon/pdfs/ContribDataMinv2.pdf&ved=2ahUKEwidntG4pMDfAhUHso8KHa28DlIQFjAAegQIARAB&usg=AOvVaw1VIefle3VfxJJiG1x7LRMA

Thank you all.

Hi

At the end of the attached paper an Illustration of the critical difference between the Connection Weight Approach (Olden and Jackson, 2002b) and Garson’s Algorithm (Garson, 1991) is given.

I hope this paper will help you.

Good luck

Isha

Same request! Would be better in matlab environment. Thanks!