Software reliability model with bathtub-shaped fault detection rate

This paper proposes a software reliability model with a bathtub-shaped fault detection rate. We discuss how the inherent characteristics of the software testing process support the three phases of the bathtub; the first phase with a decreasing fault detection rate arises from the removal of simple, yet frequent faults like syntax errors and typos; the second phase possesses a constant fault detection rate marking the beginning of functional requirements testing; the third and final code comprehension stage exhibits an increasing fault detection rate because testers are now familiar with the system and can focus their attention on the outstanding and as yet untested portions of code. We also discuss how eliminating one of the testing phases gives rise to the burn-in model, which is a special case of the bathtub model. We compare the performance of the bathtub and burn-in models with the three classical software reliability models using the Predictive Mean Square Error and Akaike Information Criterion, by applying these models to a data set in the literature. Our results suggest that the bathtub model best describes the observed data and also most precisely predicts the future data points compared to the other popular software reliability models. The bathtub model can thus be used to provide accurate predictions during the testing process and guide optimal release time decisions.