Doctor of Engineering Science
Department of Industrial Engineering and Operations Research
Columbia University
2960 Broadway, New York, NY 10027-6902, USA
For estimating the software error intensity of parametric models, weighted least-square (WLS) estimate is suggested for the first time. Unlike the widely-favored maximum likelihood (ML) estimate which only exists under certain conditions, the WLS estimate always exists as long as the expected mean and the expected variance of error count are available. The weight to the square of estimating error is chosen to be the reciprocal of the expected variance which should decrease with testing time in order to meet the intuition of weighing heavy on recently detected errors. This requirement is essentially met by the error count of RK model, but not by the most popular non-homogenous Poisson model. The WLS estimate is theoretically developed and numerically compared with least-square estimate on a set of real data. The idea of weighted estimation is also briefly discussed for other parametric
Keywords: software error intensity, non-homogenous Poisson process, weighted least-square estimates, Laplace transform, SRGM
(*Contact: E-mail slkoh.patrick@msa.hinet.net )
Cite this article as: Show-Long Patrick Koh, "Weighted Least-Square Esimate for Software Error Intensity," Journal of the Chinese Institute of Industrial Engineers, 25, 162-173 (2008).