Edinburgh Napier University

  Evolution of large-scale software systems generates very complex systems. The combination of network analysis with dynamic analysis provides a promising approach to understand such systems and support their maintenance and evolution. However, an important issue is the validity of network analysis based predictions about the functional importance of system components. Here we analyse dynamic analysis data generated for the JHotDraw 6.01b software system using network analysis methods. We show that network analysis based metrics can identify functionally important components (methods of classes) of the software system. However, we also show that some network metrics perform better than others. We show that combinations of network metrics may lead to improved performance in predicting functionally important software components, but this is again not always the case. Our results confirm the usefulness of network analysis methods in the context of dynamic analysis of software, and also underline the importance of proper validation of these methods.