Addressing the performance of two software reliability modeling methods
Peter A. Keiller, Thomas A. Mazzuchi · 2005
The problem of predicting the number of failures of a piece of software during the system test phase is addressed. Using software reliability growth models at different periods of usage of the software, predictions are made of the total number of failures one would expect at the end of the system test. Two different methods for using the models are considered: straightforward use of individual models (simple models), and dynamic selection among models based on the quality-of-prediction criteria (super models). Performance is judged by the average of the relative error of the predicted number of failures by the end of the testing period relative to the number of failures eventually observed during the interval. Three simple models and three super models are evaluated based on their performance on forty one data sets. The two methods are also investigated using smoothing techniques utilizing the Laplace trend test.