Software defect content estimation: A Bayesian approach

Abstract

Software inspection is a method to detect errors in software artefacts early in the development cycle. At the end of the inspection process the inspectors need to make a decision whether the inspected artefact is of sufficient quality or not. Several methods have been proposed to assist in making this decision like capture recapture methods and Bayesian approach. In this study these methods have been analyzed and compared and a new Bayesian approach for software inspection is proposed. All of the estimation models rely on an underlying assumption that the inspectors are independent. However, this assumption of independence is not necessarily true in practical sense, as most of the inspection teams interact with each other and share their findings. We, therefore, studied a new Bayesian model where the inspectors share their findings, for defect estimate and compared it with the Bayesian model (Gupta et al. 2003), where inspectors examine the artefact independently. The simulations were carried out under realistic software conditions with a small number of difficult defects and a few inspectors. The models were evaluated on the basis of decision accuracy and median relative error and our results suggest that the dependent inspector assumption improves the decision accuracy (DA) over the previous Bayesian model and CR models.