A challenge is presented each time to the participants at the IDRC (International Diffuse Reflectance Conference), a biennial event held in Chambersburg, Pennsylvania, USA. This is a worldwide competition with a prize for the individual or team that develops the best model and obtains the lowest prediction error for a particular dataset.
This traditional challenge is always an opportunity for learning and interacting with experienced chemometricians presenting their approach to a common multivariate analysis problem. This year, for the first time, petrochemical datasets were analysed. These were provided by Halliburton. As this is a very competitive field, steps had been taken to reduce the potential for rival companies to use the data. Specifically, the wavelength scales were unspecified, as was the nature of the parameters to be predicted. The variables corresponded to the near and mid infrared. There were two datasets to be processed, with a single parameter available per set. The first set comprised liquid samples from oil wells and the second consisted of gaseous mixtures. The spectra were acquired in transmission mode at different temperatures and pressures representing actual conditions.
The challenge was to present the best results for the validation set, for which no reference values were provided. The approach was a rather perilous one, given that the spectral space of the validation set was not covered at all by the calibration set.
The CRA-W’s NIR (Near Infrared) team thanks to its CHEMOMETRICS expertise and combining several multivariate techniques (PCA, MLR, Local-PLS, Interval -PLS and Mahalanobis distances) produced successfully the best solutions. CRA-W thus carried off the trophy for the third time out of the last five challenges, demonstrating its internationally acknowledged competence in developing spectroscopic calibration models.