Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
Context: Most research into software defect prediction ignores the differing amount of effort entailed in searching for defects between software components. The result is sub-optimal solutions in terms of allocating testing resources. Recently effort-aware (EA) defect prediction has sought to redress this deficiency. However, there is a gap between previous classification research and EA prediction. Objective: We seek to transfer strong defect classification capability to efficient effort-aware software defect prediction. Method: We study the relationship between classification performance and the cost-effectiveness curve experimentally (using six open-source software data sets). Results: We observe extremely skewed distributions of change size which contributes to the lack of relationship between classification performance and the ability to find efficient test orderings for defect detection. Trimming allows all effort-aware approaches bridging high classification capability to efficient effort-aware performance. Conclusion: Effort distributions dominate effort-aware models. Trimming is a practical method to handle this problem.