Vitenskapelig artikkel

Properties of prediction sorting

Berget, Ingunn; Næs, Tormod


Tidsskrift: Journal of Chemometrics, vol. 18, p. 92–102–11, 2004

Utgave: 2

Internasjonale standardnumre:
Trykt: 0886-9383
Elektronisk: 1099-128X

Open Access: none


One of the major sources of unwanted variation in an industrial process is the raw material quality. However, if the raw materials are sorted into more homogeneous groups before production, each group can be treated differently. In this way the raw materials can be better utilized and the stability of the end product may be improved. Prediction sorting is a methodology for doing this. The procedure is founded on the fuzzy c-means algorithm where the distance in the objective function is based on the predicted end product quality. Usually empirical models such as linear regression are used for predicting the end product quality. By using simulations and bootstrapping, this paper investigates how the uncertainties connected with empirical models affect the optimization of the splitting and the corresponding process variables. The results indicate that the practical consequences of uncertainties in regression coefficients are small. Copyright (C) 2004 John Wiley Sons, Ltd.