SINGULARITY AND PARAMETER ESTIMATION METHODS OF EXPERIMENTAL MODELS IN SOFTWARE RELIABILITY EXPERT SYSTEM (SRES)
Xu Ren · Chinese Journal of Computers · 1998
According to principle, Let failure data explain every thing, theauthors built a software reliability expert system (SRES) by adopting artificialintelligence technology. By reasoning out conclusion from fitting results offailure data of a software project, SRES can recommend users the suitablemodel as a unique software reliability measurement model. The SRES canovercome inconsistency in applications of software reliability models well. In thispaper, investigation results of singularity and parameter estimation methods ofexperimental models in SRES are reported. The experimental model base consists offourteen software reliability models, and in experimental model base of SRESthere are three software reliability models which are singular. They are non-homogeneous Poisson process model with three parameters (Nhp3ad), Littlewood'sBayesian reliability growth models(LVLM and LVQM). The singularity forms areanalyzed in paper. Further, parameter estimation methods designedspecifically for these three software reliability models are reported. At last, estimationresults by using complete software failure data, NTDS, and incomplete softwarefailure data, D. inc, are given. The calculation results demonstrate that theestimation methods are very effective to overcome singularity.