Use of robust estimators in parametric classifiersThe parametric approach to density estimation and classifier design is a well studied subject. The parametric approach is desirable because basically it reduces the problem of classifier design to that of estimating a few parameters for each of the pattern classes. The class parameters are usually estimated using maximum-likelihood (ML) estimators. ML estimators are, however, very sensitive to the presence of outliers. Several robust estimators of mean and covariance matrix and their effect on the probability of error in classification are examined. Comments are made about alpha-ranked (alpha-trimmed) estimators.
Document ID
19910035163
Acquisition Source
Legacy CDMS
Document Type
Conference Paper
Authors
Safavian, S. Rasoul (Purdue Univ. West Lafayette, IN, United States)
Landgrebe, David A. (Purdue University West Lafayette, IN, United States)