NIR spectroscopy data provide transmittance or absorbance spectrum data in the range of NIR waves. The data is 'continuous' data with peak and valley on certain range of frequency/wavelength depend on characteristics of sample. Can we apply data mining techniques for modeling or correlate this spectrum data with other physico-chemical data and how to do that?
This is commonly done with mid -Ir or Raman spectral data. To correlate with external variables, use for example Partial least squares regression.
“Multivariate statistical methodologies applied in biomedical Raman spectroscopy: Assessing the validity of partial least squares regression using simulated model datasets”,
Mark E. Keating, Haq Nawaz, Franck Bonnier, Hugh J. Byrne
Analyst, 140, 2482-2492 (2015)
I also used PLSR myself, but recently I was told about this application developed by the Africa Soil Information Service (http://africasoils.net/) that allows to analyze your mid-IR (or other type of IR spectra) with Bayesian Additive Regression Trees. I have not used it myself, but it may be worth a try. It seems it works well for them:
http://spectral.kutabiri.com/
Mercedes is correct, Bayesian approaches seem to be gaining popularity, but looking at the link quickly, it seems to be an online resource specifically for analysis of soils. The background seems quite complex. Matlab or R have a tootbox for multivariate analysis - or if you are using R there is a package available which covers alot of routines for spectral analysis http://hyperspec.r-forge.r-project.org/
Thank's for the insight! Here, we are working with fluorescence spectrum data that include intensity of exicitation and emmission spectra. Actually I still little bit confuse to differentiate between raman and fluorescence spectra data. Can R handle also fluorescence spectrum data?
Depending on the material, Fluorescecne is broadband, and derives from the radiative relaxation from an excited state. Its energy (wavelength) position is largely independent of excitation. Raman derives from coupling with the vibrational states of the material. It results in sharp bands which are characteristic of the material, and doesn't require excitation to an electronic state, although it can be enhanced resonantly. Fluorescence is normally much simpler and analysis normally doesn't require multivariate statistics... Raman spectra have much more information and therefor multivariate analysis can indicate which peaks are correlated woth different external variables. R is simply an Open Access platform, and so should deal with any data.
for very useful tutorials have a look here https://www.youtube.com/user/QualityAndTechnology
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining