(301r) Automating the Analysis of Preprocessing Techniques for Chemometrics

Preprocessing of data is often performed to improve prediction accuracy when chemometrics is used to model complex data sets. There are general rules on the selection of preprocessing techniques, but most decisions are made by technical experts. Because of the large number of possible combinations, only a small portion of the possible pretreatments is usually tested before a solution is selected. In this work we describe the use of software for the testing of thousands of preprocessing combinations. Because of the large numbers of options considered, a rapid method for the statistical evaluation and comparison of models is described. This method allows for a more thorough and systematic selection process for a final pretreatment combination for chemometric models. The development of a model for the prediction of constituent concentration based on on-line NIR data will be described as a demonstration of the method.