Abstract:

In order to improve the accuracy and stability of software reliability prediction, a method of software reliability prediction based on combination model is proposed. To build a combination model, several suitable component models should be selected, and every component model should be assigned an appropriate weight. Several criteria should be followed to select component models, which are life cycle phase, demand of model user, type of fault data, similarity of fault trend, fit of model assumptions and bias of model prediction, and an algorithm for selecting component model is gave. The sequence likelihood ratio is used to determine the weight of component model. Finally, a group of software fault data is used to check the prediction quality of the combination model. Compared with a single model, the result of the combination model is more credible.