Preliminary classification of mass spectral patterns using a simplified learning machine

Abstract

A simplified learning machine technique is presented which is suitable for forming preliminary classification of patterns. This technique allows preliminary compression of data prior to the training process, and generates a reliable classifier even when there are linearly inseparable data in the training set. This method has been used to form an eight-feature pattern classifier which identifies, directly from their mass spectra, compounds of the structure (RO)₂P(=X)Y where R is H, Me, or Et; X is O or S; and Y is any functional group.