What is the best software tool for multivariate spectroscopic (NIR) calibration for scientists?

janskvaril's picture

Please could you recommend the best software for spectroscopic calibration?

So far I was using Unscrambler (working with data-pretreatment, selection of important variables, PCA, PLS, PCR etc. ), but now I'm looking for a software enabling more scientific approach e.g. writing a script, optimization algorithm etc., more variable selection methods, different calibration methods, time series etc.

Can the SIMCA or PLS_Toolbox be a good option?

Thank you.

russell's picture

Matlab is used as the underlying tool for much scientific work in Chemometrics.

PLSToolbox is an add-in for Matlab and will provide many of the same tools you have seen in the Unscrambler plus some that don't exist in Unscrambler.

You can then write Matlab scripts to take advantage of functions in PLSToolbox and then customize with work of your own.


Dave Russell


td's picture

Hello  Janskvaril,

If you do not have access to MATLAB the "R" package might be useful. See NIR news 25(2), 19(2014). for an introduction.

Good luck,

Tony Davies


hlmark's picture

Jamslvaril - Tony failed to mention one ENORMOUS advantage of R: it's free!

The other side of that coin (the downside), however, is that as far as I know, there are no third-party software packges (like Unscrambler or PLS Toolbox) to implment the main calibration algorithms and data transforms, etc., you're interested in. You may be able to find some of them pre-written, but you would likely, at the very least, have to assemble your own "package", and you might have to code some of the algorithms yourself, too.





rogerjm's picture


another alternative is using scilab ( with fact toolbox (

it is totally free





JM Roger
Irstea, Montpellier France

ke's picture

Before this discussion devolves completely into a headless hunt for FREEWARE  allow the requested "more scientitifc" approach to variable selection etc - might I venture a word of caution. The philosophy behind The Unscrambler, and indeed the other mentioned multivariate calibration programs are actually much alike when it comes to recommending (and executing) procedures to conduct data analysis in a scientific way. But it is emphatically not just about finding THE algorithm, it is just as much about understanding: sampling, model stability, correct validation, interpretation of the underlying chemistry, outlier detection and "causality vs. indirect correlation" are important keywords here. Whatever the data analysis goals, most multivariate methods will give similar results if the model is validated in a correct way (this is NEVER cross-validation), also taking into account specific uncontrolled variation such as batch/raw material, season, instrument etc. Variable selection may sound intriguing at first glance, but does in general not give significantly better prediction ability. Testing 100's of models for the "best" one may easily lead to over-optimistic results (especially if tested by cross-validation) and, sorry to say, this is not a scientific approach. THAT said it is of course always good to remove variables that are definitely NOT of interest, either with a proper variable selection procedure and/or making use of the all-important background knowledge of the system you are dealing with (NEVER trust an algorithm without this kind of "ground truth" understanding). Some references that may be of interest:

Westad, F. & Martens, H. (2000) 'Variable selection in near infrared spectroscopy based on significance testing in partial least squares regression' JNIRS, vol 8, no. 2, pp. 117-124.

"Validation of chemometric models – A tutorial" Frank Westad, Federico Marini

Esbensen, K. & Geladi, P. (2010) "Principles of Proper Validation: Ue and absue of resampling for validation". J. Chemomtr. vol 24, p. 168-187