17 - Proportional hazards modelling of software failure data

Abstract

This chapter presents a proportional hazards model for software failure data. The proportional hazards model has had many diverse and successful applications in hardware reliability studies. This success, coupled with the, as yet limited, experience from applications in software studies, indicates that the proportional hazards model has major potential for the successful modeling of software failure data. The proportional hazards modeling technique makes few assumptions, has modeling flexibility, is particularly useful for analyzing sparse non-homogeneous data sets and naturally incorporates features known to affect software reliability that are often omitted from the classical software reliability models. The conventional software prediction models are of simple construction and do not allow for explicit modeling of these factors that affect the failure process. The limitations of the software reliability models have inevitably resulted in the introduction and applications of other techniques that are less assumption-based and/or better able to exploit the potential information in the data.