pattern recognition; bayes; maximum likelihood; mass spectrometry; proteins; bayésien; maximum de vraisemblance; spectrométrie de masse; protéines
An investigation of the properties of 3 classifiers was performed; 2 parametric and 1 non-parametric. The former classifiers were the linear and quadratic discriminant functions for which both bayesian and maximum likelihood learning (estimation) were computed; while the latter was the k-nearest neighbour classifier (k = 3 in this study). The convergence of the estimates was reported along with the classification accuracies of all classifiers. It was seen that the particular class structure studied was quite complex, leading to difficulties in predicting all samples correctly.