Table 3 Parameters and software implementations of the classification algorithms

SVM, Linear default

C (penalty parameter)

1

SVM, Linear optimized

C (penalty parameter)

optimized over (0.01, 0.1, 1, 10, 100)

SVM, Polynomial

C (penalty parameter)

optimized over (0.01, 0.1, 1, 10, 100)

q (polynomial degree)

optimized over (1, 2, 3)

SVM, RBF

C (penalty parameter)

optimized over (0.01, 0.1, 1, 10, 100)

γ (determines RBF width)

optimized over (0.01, 0.1, 1, 10, 100)/number of variables

KRR, Polynomial

λ (ridge)

optimized over (10^-10, 10^-9, …, 1)

q (polynomial degree)

optimized over (1, 2, 3)

KRR, RBF

λ (ridge)

optimized over (10^-10, 10^-9, …, 1)

γ (determines RBF width)

optimized over (0.01, 0.1, 1, 10, 100)/number of variables

KNN, default K = 1

K (number of neighbors)

1

KNN, default K = 5

K (number of neighbors)

5

KNN, optimized

K (number of neighbors)

optimized over (1, …, 50)

PNN

σ (spread)

optimized over (0.01, 0.02, …, 1)

L2-LR, default

C (penalty parameter)

1

L2-LR, optimized

C (penalty parameter)

optimized over (0.01, 0.1, 1, 10, 100)

L1-LR, default

C (penalty parameter)

1

L1-LR, optimized

C (penalty parameter)

optimized over (0.01, 0.1, 1, 10, 100)

BLR, Gaussian priors

v (variance)

automatically determined in the software by cross-validation

BLR, Laplace priors

v (variance)

automatically determined in the software by cross-validation

RF, default

ntree (number of trees)

500

mtry (number of variables sampled at each split)

$\sqrt{\mathit{number}\mathit{of}\mathit{variables}}$

RF, optimized

ntree (number of trees)

optimized over (500, 1000, 2000)

mtry (number of variables sampled at each split)

$\mathit{optimized}\mathit{over}\left(0.5,1,2\right)\times \sqrt{\mathit{number}\mathit{of}\mathit{variables}}$