ROBUST AND FAST ESTIMATION OF COVARIANCE MATRIX

Multivariate data analysis is a very important topic in nearly all fields nowadays. Principal components analysis, factor analysis, control charts are only a few multivariate methods for outlier detection, statistical process control or fault detection. Nearly all of these methods require the estimation of covariance matrix as it describes multivariate relations of all variables based on empirical moment estimation, which is very sensitive to outliers. A fast and very robust covariance matrix estimation method based on MAD (Median Absolute Deviation) will be shown, which is not only limited for Gaussian distributed data like other state of the art methods.