Unlabelled spectral data: use PCA to find distint spectra I have many (Raman) spectra which are unlabelled. I would like to use PCA to distinguish them, i.e. say all these spectra are of type A, these other spectra are of type B... etc...

*

*Does this make sense?

*Do I have to do some preprocessing like normalizing the data, if yes how?

 A: *

*If your spectra are from pure substances or well-defined mixtures (so that each mixture becomes a substance in your application):

*

*If you want to group your spectra into previously unknown groups or clusters, you want to do cluster analysis.


*If you already know what materials you want to detect, look into classification (including one-class classification).




*If your spectra can contain all kinds of mixtures of your substances, regression models are likely what you need.


*Finally, if you have mixtures but also pure substance spectra and the task is to find the pure substance spectra within the data set, spectral unmixing algorithms do that.

PCA may serve as pre-processing step for any of these methods.
Further pre-processing steps typicall need to be deterimend based on detailed knowledge about your application, how you collected the spectra (possibly know characteristics of your spectrometer setup) and the type of modeling that is to follow. It is nothing we can answer here from what you wrote.
