Considerations on the Interpretation of Mass Spectra via Learning Machines

The application of learning machine techniques to the interpretation of mass spectra is investigated. An attempt to improve the characteristic ratio of the number of training points to the number of adjustable parameters is made by the use of reduced spectra. The ensuing results are corrected for the unequal representation of the different substances in the data bank. Reasonable performance is obtained for spectra of simple (monofunctional) substances, but predictive abilities are poor for more complicated substances.