ScienceAsia 37(2011): 355-359 |doi:
New statistics for detection of outliers using the last few principal components
Rungrawee Amnarttrakul*, Ampai Thongteeraparp
ABSTRACT: Two test statistics are proposed for detecting outliers that are not extreme on any of the original variables in multivariate data. The test statistics are derived via principal component analysis and some normal distribution properties. Moreover, the last few principal components or minor principal components from principal component analysis have an important role in these simple test statistics. Finally, the tests are applied to a simulated data and the real data of a financial institution in Thailand as examples. Moreover, a comparative study was carried out using data on milk from Daudin, Duby, and Trecourt [Statistics, 19, 241].
|Statistics Department, Kasetsart University, Bangkok
* Corresponding author, E-mail: firstname.lastname@example.org
Received 11 Jul 2011, Accepted 10 Nov 2011