Improving Prediction Robustness From Hard And Soft Modeling

Dr Peter King, Managing Director of Jplus Consulting, says “Very early on we started a dialog with CAMO Software who recognized the complementary nature of our ‘hard modeling’ approaches to the classical chemometric multivariate analyses such as PCA and PLS and ‘soft modeling’ techniques included in The Unscrambler® X.”

Essentially both approaches analyse chemical data to gain information about the underlying process or system using mathematical and statistical tools. The key difference is that the chemometric algorithms do not require a starting model but allow the development of an abstract model from patterns and behavior in training samples which can then be used to predict the constitution and key characteristics of unknown samples. These characteristics may not be ‘real’ parameters such as absorption spectra or concentration profiles but abstract mathematical decompositions, which nonetheless allow very powerful prediction of unknowns. "The Unscrambler® X is very well equipped in this area and offers a wide range of tools in an exceptionally user friendly environment” says Peter.

‘Hard modeling’ starts with a mathematically rigorous model of the process under investigation. In the case of ReactLab, this is the reaction scheme underlying a chemical process. This model is then fitted to the measurement. This involves the adjustment and optimisation of the key rate and equilibrium parameters using a least squares approach. This yields the fundamental reaction parameters that define the process which can be re-used in more complex systems and by other researchers.