Making the best prediction from your DOE test

[#1]

Hi all, 

I've just finish an DOE test with 3 factors & the result is only 1 factors is statistically significant a.k.a having impact to the response. (meaning the remain factors have no impact/ too small impact vs system noise)

- Some pre-study & DOE result show that the noise is ~0.3

Here is the prediction profiler

I'm trying to use this result to predict the response & I'm considering 2 options that I don't know which one is the right way

- Option 1: Using DOE's profiler & predict by adding noise from system. Meaning I use the mean from predictive formular & add a random noise

- Option 2: I use the confident interval of response & adding noise from system. I create a new profiler when add the range of CI & add noise. (i'm thinking this one is better predict the real system)

So can anyone tell me which way is correct in my way. 

And are there any better way can I do? 

Thanks alot 

[#2]

Unfortunately I don't understand your situation enough to provide good advice. What questions are you trying to answer with your study?  Are you interested in how well the model predicts the actual results of the process in the future?  Do you want to improve on the model's ability to predict the future? There is no one way to use your data for prediction.  We are missing some pertinent information (e.g., the Design and how the experiment was run, RSquare-RSquare Adjusted).  What is meant by "Some pre-study & DOE result show that the noise is ~0.3". Is 0.3 an estimate of the standard deviation of the response variable?  Is that consistent over time? Is that a lot or is that small (I have no context).

Here are my thoughts:

1. First, ignoring statistical significance, was there a practically significant change in the response variable?  This can be answered by the SME.  

2. Did any of the factors exhibit a practically significant change in the response variable?

3. Have you identified what the noise was during the experiment?  How representative is the noise that changed during the experiment of future conditions? 

4. For the factors that were insignificant, how bold was the level setting?  Were the levels balanced?

5. Did you save and plot the residuals?  Do they meet the NID(mean, variance) assumptions?

Again, I don't know your situation, but you might start with the saved prediction formula and run the process "over time" and evaluate the accuracy and precision of the model and subsequently evaluate the residuals.  Or try each of your described methods and see which one gives better results?