MULTIVARIATE SELECTION OF VARIABLES IN INDUSTRIAL QUALITY-CONTROL - OPTIMIZING AVIATION FUEL FINAL CONTROL

This paper deals with a typical question encountered in all industrial analytical laboratories: are all the tests performed in the laboratory to characterize the final product really necessary? In this work the cross-validation technique, Procrustes rotation techniques and statistical variable selection have been used to reduce the 26 initial British Petroleum and ASTM kerosene specification tests to ten 'essential' ones. Statistical as well as chemical considerations were used to select the optimum subset of original variables to be retained from all the possible ones.