I-Optimal Design is an optimal experimental design that minimizes the average prediction variance across the entire design space — focusing on the accuracy of predictions rather than parameter estimates, making it the preferred criterion when the goal is to build a predictive model.

I-Optimal vs. D-Optimal

• I-Optimal: Minimizes $int Var[hat{y}(x)] dx$ (integrated prediction variance over the design space).
• D-Optimal: Minimizes parameter variance (maximizes $|X^TX|$).
• For Prediction: I-optimal produces better predictions on average; D-optimal produces more precise parameters.
• Software: JMP and other DOE software support I-optimal design generation.

Why It Matters

• Surrogate Models: When the goal is building a predictive model (virtual metrology, response surface), I-optimal is the best criterion.
• Process Optimization: Better predictions lead to more accurate identification of the optimal operating point.
• Design Space: I-optimal designs typically place more points at the boundaries of the factor space.

I-Optimal Design is designing for the best predictions — minimizing prediction error across the entire design space for the most accurate process model.