Building structure-property predictive models using data assimilation

Abstract

In Chemometrics it is often the norm to develop regression methods for analysing non-linear multivariate data by using the observations (measurements) as the sole constraint. This is the case regardless of the nature of the regression method (parametric or non-parametric) [1]. In this article we present the development of a regression model using data assimilation[2] - A technique that takes into account additional available information about the "system" which the model is to represent. The new approach shows substantial improvement over the "conventional" methods[3] against which it has been compared. © Springer-Verlag Berlin Heidelberg 2006.