An empirical approach to update multivariate regression models intended for routine industrial use

Many problems currently tackled by analysts are highly complex and, accordingly, multivariate regression models need to be developed. Two intertwined topics are important when such models are to be applied within the industrial routines: (i) Did the model account for the ‘natural’ variance of the production samples? (ii) Is the model stable on time? This paper focuses on the second topic and it presents an empirical approach where predictive models developed by using Mid-FTIR and PLS and PCR hold its utility during about nine months when used to predict the octane number of platforming naphthas in a petrochemical refinery.