The signal processing of spectroscopic data is a well-researched topic area. You will find thousands of papers written on the topic in a Google Scholar search.

The signal processing of spectroscopic data is a well-researched topic area. You will find thousands of papers written on the topic in a Google Scholar search. Signal processing, which is a domain within the electrical engineering discipline, is a good place to start your reading and learning. ML is a poor fit in my opinion for spectroscopic analysis because the underlying physics is absolute: every element or molecule has one and only one signature within a given band, and it is typically at a very high SNR. Now, it is possible that your sensor(s) can have variability so that is the only case where I see a ML vs. model-based approach (e.g. matched filter) being applicable. Good luck on your journey.