Elastic Net

models = "ENet"

Type: Classification, Regression

Tuning parameter defaults:

  • lambda = 1
    • The quadratic penalty parameter.

Required packages: elasticnet

K Nearest Neighbors

models = "KNN"

Type: Classification, Regression

Tuning parameter defaults:

  • k = 10
    • The number of nearest neighbors to consider.

Required packages: class

Lasso

models = "Lasso"

Tuning parameter defaults:

  • lambda = .2
    • The shrinkage.

Type: Classification, Regression

Required packages: lars

Least Angle Regression

models = "LARs"

Tuning parameter defaults:

  • lambda = .05
    • The shrinkage.

Type: Classification, Regression

Required packages: lars

Neural Networks

models = "NNet"

Type: Classification, Regression

  • size = 2
    • The number of neurons in the hidden layer.
  • decay = 0
    • Weight decay. Controls the regularization of the cost function.

Required packages: nnet

Partial Least Squares Linear Discriminant Analysis

models = "PLSLDA"

Type: Classification

Tuning parameter defaults:

  • ncomp = min(floor(n/nfolds), p, 100))
    • The number of components. n is the number of observations. nfolds is the number of folds. p is the number of parameters.

Principal Components Regression

models = "PCR"

Type: Classification, Regression

Required packages: pls

Random Forest

models = "Forest"

Type: Classification, Regression

Tuning parameter defaults:

  • mtry = ifelse(classify, max(floor(p/3), 1), floor(sqrt(p)))
    • Number of variables randomly sampled as candidates at each split. p is the number of descriptors.

Required packages: randomForest

Recursive Partitioning (RPart)

models = "RPart"

Type: Classification, Regression

Tuning parameter defaults:

  • cp = .01
    • The overall R-squared must increase by cp with each split.

Required packages: rpart

Recursive Partitioning (Tree)

models = "Tree"

Type: Classification, Regression

Required packages: tree

Ridge Regression

models = "Ridge"

Type: Classification, Regression

Tuning parameter defaults:

  • lambda = .1
    • The shrinkage.

Required packages: MASS

Support Vector Machines

models = "SVM"

Type: Classification, Regression

Notes

Depending on whether y is binary or continuous, C-classification or eps-regression is used.

Tuning parameter defaults:

  • gamma = 1
    • The gamma controls how local/flexible the fit is (the larger, the more local the fit)
  • cost = 1
    • For C-classification, controls the number and the severity of the violations of the margin in maximal margin classification. The larger C is, the more “budget” for violations.
  • epsilon = .01
    • For epsilon-Regression, epsilon controls what is comparable to the size of the margin in C-classification

Required packages: e1071