Deep learning model fitting parameters for best fit in Knowledge studio

I am trying to fit two independent variable dataset to predict a dependent variable in a deep learning neural network. After prediction, I am trying to do a log-log plot of original and predicted value against one of the independent variables. It looks like my assumed neural network parameters of deep learning like number of hidden layers = 4 , number of nodes per hidden layer = 10 , maximum iteration = 10,000, BFGS optmizer with grad epsilon value = 1E-10 and line epsilon value = 0.1, fits well for mapping log values above 1 but it has worse fit for log values below 1. What would you suggest to change in my neural network parameters to predict more accurately?