Chapter 4. Fundamentals of design of experiments and optimization: data modeling in response surface methodology

CHIAPPINI, FABRICIO A.; AZCARATE, SILVANA; TEGLIA, CARLA; GOICOECHEA, HÉCTOR C.

Introduction to Quality by Design in Pharmaceutical Manufacturing and Analytical Development

Springer

Año: 2023; p. 30 - 39

The experimental data collected according to a proper statistical design constitute the input for data modeling, which represents the last step in response surface methodology (RSM). Data modeling consists in applying a set of statistical methods that enable the analyst to thoroughly study the relation between experimental factors and responses. Essentially, this task is carried out by building empirical models, which are then used to make predictions and investigate possible optimal experimental regions. In this chapter, important concepts of multivariate statistics, related to data modeling in RSM are introduced. In particular, models based on multiple linear regression, MLR (parametric) and artificial neural networks, ANN (non-parametric) are presented, which are the two most important modeling approaches in RSM. Additionally, relevant issues regarding model validation, outlier diagnosis, prediction and interpretation are discussed, and mathematical methods for single and multiple response optimization are briefly described. Finally, some of the most popular software for RSM implementation are summarized.