slda {ipred} R Documentation

## Stabilised Linear Discriminant Analysis

### Description

Linear discriminant analysis based on left-spherically distributed linear scores.

### Usage

```## S3 method for class 'formula':
slda(formula, data, subset, na.action=na.rpart, ...)
## S3 method for class 'factor':
slda(y, X, q=NULL, ...)
```

### Arguments

 `y` the response variable: a factor vector of class labels. `X` a data frame of predictor variables. `q` the number of positive eigenvalues the scores are derived from, see below. `formula` a formula of the form `lhs ~ rhs` where `lhs` is the response variable and `rhs` a set of predictors. `data` optional data frame containing the variables in the model formula. `subset` optional vector specifying a subset of observations to be used. `na.action` function which indicates what should happen when the data contain `NA`s. Defaults to `na.rpart`. `...` additional parameters passed to `lda`.

### Details

This function implements the LDA for q-dimensional linear scores of the original p predictors derived from the PC_q rule by Laeuter et al. (1998). Based on the product sum matrix

W = (X - bar{X})^top(X - bar{X})

the eigenvalue problem WD = diag(W)DL is solved. The first q columns D_q of D are used as a weight matrix for the original p predictors: XD_q. By default, q is the number of eigenvalues greater one. The q-dimensional linear scores are left-spherically distributed and are used as predictors for a classical LDA.

This form of reduction of the dimensionality was developed for discriminant analysis problems by Laeuter (1992) and was used for multivariate tests by Laeuter et al. (1998), Kropf (2000) gives an overview. For details on left-spherically distributions see Fang and Zhang (1990).

### Value

An object of class `slda`, a list with components

 `scores` the weight matrix. `mylda` an object of class `lda`.

### Author(s)

Torsten.Hothorn <Torsten.Hothorn@rzmail.uni-erlangen.de>

### References

Fang Kai-Tai and Zhang Yao-Ting (1990), Generalized Multivariate Analysis, Springer, Berlin.

Siegfried Kropf (2000), Hochdimensionale multivariate Verfahren in der medizinischen Statistik, Shaker Verlag, Aachen (in german).

Juergen Laeuter (1992), Stabile multivariate Verfahren, Akademie Verlag, Berlin (in german).

Juergen Laeuter, Ekkehard Glimm and Siegfried Kropf (1998), Multivariate Tests Based on Left-Spherically Distributed Linear Scores. The Annals of Statistics, 26(5) 1972–1988.

`predict.slda`

### Examples

```
learn <- as.data.frame(mlbench.twonorm(100))
test <- as.data.frame(mlbench.twonorm(1000))

mlda <- lda(classes ~ ., data=learn)
mslda <- slda(classes ~ ., data=learn)

print(mean(predict(mlda, newdata=test)\$class != test\$classes))
print(mean(predict(mslda, newdata=test)\$class != test\$classes))

```

[Package ipred version 0.8-1 Index]