Model-based software reliability prediction

We investigate an approach for early estimation of the reliability of software products based on their design models. We start by computing several structural and behavioral metrics from Unified Modeling Language(UML) models. The choice of the selected metrics has been based on their ability to influence the final product reliability. A trained neural net is used to predict the reliability of individual modules. The final product reliability is obtained from these predicted values. Our approach can help to decide between design alternatives and also help a manager trade off between the cost of redesigning certain modules and increased testing effort to meet product reliability goals. abstract environment. © 2010 Springer-Verlag Berlin Heidelberg.