A dynamically-weighted software reliability combination model

In this paper, a dynamically-weighted software reliability combination model is proposed, aiming at improving predictive quality. The process of forming the combination model includes selection of component models and determination of weight assignment. Criteria for selecting component models include life cycle phase, trend exhibited by failure data, validity of model assumptions and predictive biases of model. The method of determining weights of component models is based prequential likelihood. A set of failure data is introduced for validation, the result of which implies that the prediction given by the combination model is more dependable than single model.