KNN classification - evaluated by repeated double cross validation: Recognition of minerals relevant for comet dust

Repeated double-cross validation (rdCV) has recently been suggested as a careful and conservative strategy for optimizing and evaluating empirical multivariate calibration models. This evaluation strategy is adapted in this work for k-nearest neighbor (KNN) classification. The basics of rdCV are described, including the search for an optimum k, and tests with Italian Olive Oil Data. KNN-rdCV is applied to classify 17 mineral groups, relevant for the composition of comet dust particles, characterized by the peak heights at 20 selected masses in time-of-flight secondary ion mass spectra (TOF-SIMS). Predictive abilities for 15 mineral classes are >95%, for two classes 75 and 85%. (C) 2014 Elsevier B.V. All rights reserved.