| Popis: |
Singular value decomposition (SVD) of rectangular datasets has proved to be a useful multivariate data decomposition approach because of its ability to decompose both square and rectangular matrices. Choi & Huh (1996) SVD approach utilizes the median as a robust location estimate other than the mean estimate used in the ordinary SVD. Since insufficient dataset and in addition, inefficiency of the median estimate seems to be a major setback of the existing robust SVD systems, this study envisaged an integrated algorithm that incorporated the BootSVD of Fisher (2016) and the sample Myriad estimate in cropping a new SVD system, the Robust Bootstrapped SVD (RobBootSVD). The new RobBootSVD was appraised alongside the existing ones using the Principal Component Analysis biplot quality measures and the new RobBootSVD T2 measure. Applications on tobacco process datasets with both short and long runs and simulated datasets with various percentages of outliers showcases the viability of the new approach. JEL Classification: C11, C15, C19, C55 |
