Fast-tracking design space identification with the prediction reliability enhancing parameter (PREP)

• Introduced the Prediction Reliability Enhancing Parameter (PREP) for improving prediction reliability in latent variable modeling. • PREP accelerates convergence to the True Design Space (TDS) by efficiently identifying high-reliability samples. • Validated PREP on nonlinear datasets, achieving targets with fewer iterations than conventional methods. In industrial product development, latent variable modeling tools are widely used to address challenges like multicollinearity and small sample sizes. However, these methods are often limited by prediction uncertainty, particularly when identifying optimal operating conditions or formulations to achieve desired product characteristics. This study introduces a methodology that leverages latent variable modeling alignment metrics, including partial least squares and principal components analysis Hotelling T², Sum of Squared Prediction Errors (SPE), and score alignment metrics (h PLS and h PCA ), to quantify and enhance prediction reliability. These metrics are integrated into a Prediction Reliability Enhancing Parameter (PREP), a quantitative measure designed to identify recipes with higher reliability relative to the general model uncertainty. Using an iterative optimization-based algorithm, the methodology expands the Knowledge Space (KS) to efficiently determine the True Design Space (TDS), even when the TDS lies outside the KS. Validation with simulated nonlinear datasets demonstrates that the PREP approach achieves desired targets with significantly fewer iterations compared to conventional methods, particularly in cases in which the data are highly non-linear. The PREP approach thus provides a practical and effective solution for improving prediction reliability in complex, data-driven product design, offering enhanced accuracy and flexibility in identifying optimal formulations or operating conditions.