Random Forests, Nearest Shrunken Centroids and Support Vector Machines for the Classification of Diverse E-Nose Datasets.