The Panga (Kaji) Competition

You are invited to enter the chemometrics competition. A similar competition has been a feature of the International Diffuse Reflectance Conference at Chambersburg, USA for some years.

A set of NIR spectral data is available for downloading from the ANISG website and contestants are asked to find and explain a chemometric model to robustly predict samples of the same type.

The Data

Within the zip there are csv files containing 500 samples (labelled Samples 1:500) for two different NIR instruments A and B (each instrument has its own csv file). There is then another 500 samples (labelled Samples 501:1000) which is for one of the instruments (Instrument A) along with accompanying reference results (Crude Protein, CP, across two different csv files).  There are also 2 different test sets (csv files), each of 100 samples from both instruments.

The Challenge

To assess the integrity of the data and then build a robust calibration and then predict CP results for both test sets for both instruments. The reference values for these samples are known, but not given as this is what is being assessed. The predictions will need to be submitted before presentation of the challenge for calculation of summary performance statistics (i.e. RMSEP) by the judges.

Contestants will need to build a calibration from Instrument A data that predicts the results, and then build a transformation / calibration from Instrument B to Instrument A.

The test set for Instrument A can be applied directly to the reference calibration, whereas the Instrument B spectra will first need to be transformed to look like Instrument A data to be used with the reference calibration.

Contestants will need to present their approach to the problem and discuss algorithms and solutions used in 5 or less PowerPoint slides.

There will be two prizes for the challenge, one for the best calibration for test set instrument A, and one for test set Instrument B. ‘Best’ will be judged both on the performance of the test sets and the approaches taken.

Feel free to only attempt the first part of the challenge, building a robust calibration for the reference result for Instrument A, and only submitting test set Instrument A if you wish.

Any questions or clarifications feel free to contact Kyle at

Happy calibration building!

Download the Data