do you mean R squared of each individual input with respect to the output?

a brute force way to do this is pass features step wise to the NN, and compute R sq. for each built NN with respect to the output.

t = target

y = output

CorrcoefYT = corrcoef(y,t)

R = CorrcoefYT (2,1)

it is the comparison between exact values and predictive value

Dear Mohamed

yes, it is correlation between actual value and predicted one, which we know it as correlation between output and target.