Software reliability growth models for discrete and incomplete testing

A model of the numbers of faults found on successive tests of software modules undergoing development is proposed. Testing occurs only at discrete times, and may not always be completed. Testing is also assumed to be imperfect, in that some faults may not be detected. There is assumed to be an initial distribution of the number of faults per module, and faults which are found on testing are assumed to be fixed, with the possible consequence of the production of further faults in the software (imperfect debugging). A related model assumes that faults are produced initially and are detected in clusters. The likelihood functions for the models are derived, and the Empirical Bayes method used to fit the models to data on the testing of a number of software modules used to control a complex device. The model is used to predict how many faults are likely to remain in the software after testing is completed.