Table 1.

Comparison of the six classifiers (as implemented here)

	Cor pattern- correlation classifier	KNN k-nearest- neighbors classifier	LDA Fisher’s linear discriminant analysis	GNB Gaussian naive Bayes	SVM-lin linear support vector machine	SVM-RBF radial-basis- function support vector machine
type	nearest neighbor		Gaussian		SVM
decision boundary	linear	nonlinear	linear	nonlinear (quadratic surface)	linear	nonlinear
pattern distribution model	-	nonparametric	Gaussian (same for each class, correlated voxels)	Gaussian (distinct for each class, independent voxels)	-	-
related distance function	correlation distance	correlation distance	Mahalanobis distance	Mahalanobis distance	Euclidian distance	Euclidian distance
regularization	within-class averaging	training-set cross-validation to select k (defining the size of the neighborhood)	Gaussian assumption, class-pooled, optimal- shrinkage covariance estimate	Gaussian assumption ignoring voxel correlations	training-set cross-validation to select C (defining misclassification penalty)	training-set cross-validation to select C and γ (defining RBF width)