Design of experiments and artificial neural networks as useful tools in the optimization of analytical procedure.

Abstract

Developing the analytical procedure requires estimating what independent variables will be tested and at what levels. There are statistical models that enable the optimization of the process. They involve statistical analysis, which indicates the crucial factors for the process and the potential interactions between the analyzed variables. Analysis of variance (ANOVA) is applied in the evaluation of the significance of the independent variables and their interactions. The most commonly used chemometric models are Box-Behnken Design, Central Composite Design and Doehlert Design, which are second-order fractional models. The alternative may be the artificial neural networks (ANN), whose structure is based on the connection of neurons in the human brain. They consist of the input, hidden and output layer. In such analysis, the activation functions must be defined. Both approaches might be useful in planning the analytical procedure, as well as in predicting the response prior to performance the measurements. The proposed procedures may be applied for polymeric systems.