Investigating a specific class of software reliability growth models

Abstract

The performance of a subset of the software reliability growth models is investigated using various smoothing techniques. The method of parameter estimation for the models is the maximum likelihood method. The evaluation of the performance of the models is judged by the relative error of the predicted number of failures over future time intervals relative to the number of failures eventually observed during the interval. The use of data analysis procedures utilizing the Laplace trend test are investigated. These methods test for reliability growth throughout the data and establish "windows" that censor early failure data and provide better model fits. The research showed conclusively that the data analysis procedures resulted in improvement in the models' predictive performance for 41 different sets of software failure data collected from software development labs in the United States and Europe.