constraining 1 output variable

[#1]

Hi,

I am wanting to apply a constraint to one of six output variables in a prediction profiler as currently the profiler predicts positive values when the input values match those used to form the model but generates negative values when trying to extend the data to inputs not used to build the model 

Thanks

[#2]

You do not state the type of model that you are using for prediction. I will assume that it is a linear regression model.

You might use a transformed response with your model, where the transform linearizes the response and satisfies the linear regression assumption that the fitted response may vary from negative infinity to positive infinity. For example, when the response is physically constrained to [0,1] (e.g., yield 0-100%), then the logit transform of Y helps.

You might use a different type of model, such as a neural network or  a non-linear model from theory for the response curve that you desire.