Table 4.

Simulation study parameters.

Technique	Parameter	Type	Value/Search Grid
PLS-DA	Number of components	Optimized	[1, 2, ..., 15]
Sparse PLS-DA	Number of components	Optimized	[1, 2, ..., 15]
	Regularization ($\lambda$)	Optimized	[0.1, ..., 0.9] by 0.1
Random Forest	Ensemble size	Fixed	1000
	Random subspace size	Optimized	[5, ..., p] of length 25
SVM	Kernel	Fixed	Gaussian
	Bandwidth ($\mathsf{\gamma }$)	Optimized	10^[−5, ..., −1] of length 1000 †; 10^[−2, ..., 0] of length 1000 ‡
Neural Network	Number of hidden layers	Optimized	1 or 2
	Number of hidden nodes	Optimized	[15, ..., 100] by 5
	Activation function	Fixed	Logistic
	Learning function	Fixed	Resilient Backpropagation
	Error function	Fixed	Cross-entropy Loss
k-NN	Number of neighbors	Optimized	[1, 2, ..., 20]

^† Prior-to significance filtering. ^‡ Post-significance filtering.