Powered by FOSS

NIR prediction model contest at the IDRC software shoot-out

by Martin Andersson

As one of the traditional events of the 2012 International Diffuse Reflectance Conference in Chambersburg there was an NIR calibration competition. Five experienced chemometricians presented their approaches to a common multivariate analysis problem for pharmaceutical tablets targeting the active pharmaceutical ingredient Escitalopram.


Near infrared (NIR) transmittance spectra were recorded from 4000 cm–1 to 14,000 cm–1 (714 to 2500 nm), with a resolution of 16 cm–1 and 128 co-added scans per sample.
Contestants were to develop the best NIR prediction model with the available data and send their predictions of the validation set to the shoot-out chair prior to the beginning of the conference. Criteria for deciding winners included prediction statistics of the validation set, novelty and uniqueness of the approach, and clarity of the presentation.

The chemometricians chose quite similar approaches and their results were close, around 1.6 times the error of the reference method. The participant with the lowest RMSEP for the unknown sample set received most of the public votes and was selected as the winner.

Read the full article and…

• Learn more about the different approaches and find out which approach was voted highest by the audience
• Find out which participant developed the best NIR prediction model
• See which problems occurred during the contest and how the participants coped with the challenges
• And finally – find out which participant won the calibration competition!