Software Metrics Suites for Project Landscapes

Many software metrics have been proposed over the past decades. Selecting a small custom suite of metrics is desirable for quality assessment and defect prediction in industrial practice since developers cannot easily cope with dozens of metrics. In large software architectures, structurally similar projects are subject to similar defect conditions and can be analyzed by the same metrics suite. A large Java application developed at Continentale Insurance contains structurally similar subprojects for different insurance branches (health, life, accident, car, property). These branches are integrated in the IT-architecture in a technically uniform way. We investigated which subsets of metrics are predictive but uncorrelated with each other and compared the results for structurally similar projects.