A new approach to the classification of igneous rocks using the basalt-andesite-dacite-rhyolite suite as an example

The obvious need to be able to predict, on a routine basis, the name of an igneous rock from its chemistry is examined in the light of some multivariate statistical methods. Discriminant analysis is ruled out for general use, due to its inability to distinguish an unknown that does not belong to any of the groups set up by the analysis, and the complexities of the decision-making logic with large number of groups. The best method appears to be classification procedure which predicts the probabilities of an unknown belonging to a particular group. The two methods are illustrated using samples of basalts, andesites, dacites and rhyolites.