Software reliability prediction improvement with prior information incorporated

Abstract

Software reliability growth models (SRGMs) play a central role in reliability prediction, providing critical measurements for decision makings in software testing. Against the immediate fault correction assumption, extensions on traditional NHPP models have been studied by associating fault correction process (FCP) with fault detection process (FDP). However, as they are applied by fitting against historical data, such paired models for both FDP&FCP cannot provide accurate estimations in the early phase of testing, similar to traditional NHPP models. Only in the later phase can accurate estimation be made, but then sometimes it would be too late to be useful. It is necessary to estimate accurately in the early phase for timely decision-makings. Stepping from this point, this paper studied the improvement of reliability prediction by incorporating information from prior similar project or release, within the context of paired FDP&FCP models. Limited by the available datasets, this study is carried out with simulation. The paired modeling framework provides a convenient simulation approach. Improvement in reliability prediction is observed with the analysis on simulated datasets. Also, with a cost model incorporating the factor of fault correctors' number, the optimal release time is discussed and illustrated in a simulated approach.