| Abstrakt: |
The classification of low-resolution mass spectral data is one of the oldest and most debated problems in the chemical applications of pattern recognition. This paper reviews the problems and applies all the major preprocessors and classifiers suggested by previous workers, along with some that have not been applied to mass spectral data, to single, well- defined database. The effects of six different preprocessors, six nonprobabilistic classifiers, and two forms of the first chemical probabilistic classifier are studies and their implications considered. The autocorrelation preprocessor and the k-nearest-neighbour classifier are found to be superior. [ABSTRACT FROM AUTHOR] |
