GIMCD: Gaussian imputation followed by MCD

In modi: Multivariate outlier detection and imputation for incomplete survey data

Description

Gaussian imputation uses the classical non-robust mean and covariance estimator and then imputes predictions under the multivariate normal model. Outliers may be created by this procedure. Then a high-breakdown robust estimate of the location and scatter with the Minimum Covariance Determinant algorithm is obtained and finally outliers are determined based on Mahalanobis distances based on the robust location and scatter.

Usage

GIMCD(data, alpha = 0.05,  seedem, seedmcd)

Arguments

data	a data frame or matrix with the data
alpha	a threshold value for the cut-off for the outlier Mahalanobis distances
seedem	random number generator seed for EM algorithm, default is 234567819
seedmcd	random number generator seed for MCD algorithm, if seedmcd is missing an internal seed will be used.

Details

Normal imputation from package norm and MCD from package MASS. Note that currently MCD does not accept weights.

Value

Result is stored in a global list GIMCD.r:

center	robust center
scatter	robust covariance
alpha	Quantile for cut-off value
computation.time	Elapsed computation time
outind	logical vector of outlier indicators
dist	Mahalanobis distances

Author(s)

Beat Hulliger

References

B\'eguin, C. and Hulliger, B. (2008) The BACON-EEM Algorithm for Multivariate Outlier Detection in Incomplete Survey Data, Survey Methodology, Vol. 34, No. 1, pp. 91-103.

See Also

cov.mcd, norm

Examples

data(bushfirem)
det.res<-GIMCD(bushfirem,alpha=0.1)
print(det.res$center)
PlotMD(det.res$dist,ncol(bushfirem))

Example output

GIMCD has detected 14 outliers in 0.14 seconds.      X1       X2       X3       X4       X5 
109.4322 149.1318 253.9140 212.7733 275.2639 
$hmed
[1] 21.87818

$halpha
      35 
272.1149 

modi documentation built on May 2, 2019, 6:48 p.m.