A Semiparametric Software Reliability Model for Analysis of a Bug-Database With Multiple Defect Types

Software bug-databases provide an important source of data for assessing the reliability of a software product after its release. Statistical analysis of these databases can be challenging when software usage is unknown, that is, there is no information about the usage, either in the form of a parametric model, or in the form of actual measurements. Reliability metrics, when defined on a calendar time scale, would depend on this unknown and time-dependent usage of the software and hence cannot be estimated. This article proposes a semiparametric analysis that makes use of defect classifications into multiple types to enable estimation of a model without making strict assumptions about the underlying usage of the software. New reliability metrics whose computation does not depend on the unknown usage of the software have been proposed and methods for estimating them have been developed. The proposed method has been illustrated using data retrieved from the bug-database of a popular scripting language, named Python. This illustration compares reliability of two versions of the software without making assumptions about their unknown usage. This article has supplementary material online.