SOFTWARE RELIABILITY MODEL SELECTION: A NEW LOOK ON THE U-PLOT METHOD

摘要

The U-plot method is one of the most famous statistical methods for assessing the fit of software reliability models on real data. This method was initially presented simply as a graphical tool for the selection of appropriate models. In this paper, we first show that the U-plot method can be considered as a statistical goodness-of-fit test for the Power-Law Process (or Duane model). A Monte-Carlo simulation study shows that this result is also probably true for the Moranda Geometric and Goel-Okumoto models. Finally, it is proved that the choice of the Anderson-Darling instead of the Kolmogorov-Smirnov distance improves the efficiency of the method. The application to real data shows that this method helps prac-titionners to select more appropriate models.