calibrated multivariate outlier detection

interface to a parametric multivariate outlier detection algorithm

mv.calout.detect(x, k = min(floor((nrow(x) - 1)/2), 100), Ci = C.unstr, 
    lamfun = lams.unstr, alpha = 0.05, method = c("parametric", 
        "rocke", "kosinski.raw", "kosinski.exch")[1], ...)

Arguments

x	data matrix
k	upper bound on number of outliers; defaults to just less than half the sample size
Ci	function computing Ci, the covariance determinant ratio excluding row i. At present, sole option is C.unstr (Caroni and Prescott 1992 Appl Stat).
lamfun	function computing lambda, the critical values for Ci
alpha	false outlier labeling rate
method	string identifying algorithm to use
...	reserved for future use

Details

bushfire is a dataset distributed by Kosinski to illustrate his method.

Value

a list with components

inds

indices of outlying rows

vals

values of outlying rows

k

input parameter k

alpha

input parameter alpha

References

C. Caroni and P. Prescott, Journal of the Royal Statistical Society. Series C (Applied Statistics), Vol. 41, No. 2 (1992), pp. 355-364

Author

VJ Carey

Examples

data(tcost)
mv.calout.detect(tcost)
#> $inds
#> [1] 21  9
#> 
#> $vals
#>     fuel repair capital
#> 21 26.16  17.44   16.89
#> 9  29.11  15.09    3.28
#> 
#> $k
#> [1] 17
#> 
#> $alpha
#> [1] 0.05
#> 
data(bushfire)
mv.calout.detect(bushfire)
#> $inds
#> [1]  7 11 10  8  9
#> 
#> $vals
#>      [,1] [,2] [,3] [,4] [,5]
#> [1,]   92  110   46  165  235
#> [2,]  108  115   17  144  215
#> [3,]  100  104   21  133  208
#> [4,]   94   95   29  113  190
#> [5,]   94   94   29  110  188
#> 
#> $k
#> [1] 18
#> 
#> $alpha
#> [1] 0.05
#>