Abstract: The present invention is enclosed in the area of machine learning, in particular machine learning for the analysis of High or Super-resolution spectroscopic data, which typically comprises analysis of highly complex samples/mixtures of substances and/or data with low resolution, for instance Laser-Induced Breakdown Spectroscopy (LIBS). It is an object of the present invention a method of computational self-learning for characterization of one or more constituents in a sample, from electromagnetic spectral information of such sample, which changes the paradigm associated with prior art methods, by using only sub-optical spectral information, i.e., obtaining the resolution of the spectral information and thereby be able to extract spectral lines—thus determining a spectral line position—from such spectral information, hence avoiding all the uncertainty associated with pixel based methods. It is also an object of the present invention a computational apparatus configured to implement such method.