A Single Changepoint Software Reliability Growth Model with Heterogeneous Fault Detection Processes

Most software reliability growth models characterize the software process as a function of testing time. However, during the software testing process, the failure data is affected by additional factors such as testing strategy and environment, integration testing, and resource allocation. This will have a major impact on the fault detection process reflecting the effect of such factors at various stages of testing, which are known as changepoints. Recently, several researchers have proposed non-homogeneous Poisson process software reliability models with one or more changepoints to model the data well. However, one of the limitations of previous research is that only homogeneous combinations of failure distributions before and after changepoints are considered. However, in real data sets this is often not the case. This paper develops heterogeneous single changepoint models by considering different failure distributions before and after the changepoint and applies algorithms to maximize the likelihood of these models. Heterogeneous models are compared with existing homogeneous models using goodness-of-fit measures. The expectation conditional maximization algorithm identifies the maximum likelihood estimates of the model parameters. Online changepoint analysis is also described. Experimental results suggest that heterogeneous changepoint models better characterize some failure data sets.