Measuring software reliability with a growth curve

The number of defects found during testing rises steeply at first and then levels off. The shape closely resembles the Gompertz curve, long used to describe biological growth and the diffusion of innovations.

This work formulates the Gompertz curve as a software reliability growth model and gives an algorithm that estimates its parameters stably. Earlier methods could diverge depending on the initial values chosen; the proposed method avoids that. Validation on data from real development projects showed better predictive accuracy than existing models.

It is one of the most frequently cited results from our laboratory.