Predicting software black-box defects using stacked generalization

Abstract

Defect number prediction is essential to make a key decision on when to stop testing. For more applicable and accurate prediction, we propose an ensemble prediction model based on stacked generalization (PMoSG), and use it to predict the number of defects detected by third-party black-box testing. Taking the characteristics of black-box defects and causal relationships among factors which influence defect detection into account, Bayesian net and other numeric prediction models are employed in our ensemble models. Experimental results show that our PMoSG model achieves a significant improvement in accuracy of defect numeric prediction than any individual model, and achieves best prediction accuracy when using LWL(Locally Weighted Learning) method as level-1 model.

Automated bug assignment: Ensemble-based machine learning in large scale industrial contexts

“…One prominent example was developed by the winning team of the Netix Prize, where a solution involving SG outperformed the competition in predicting movie ratings, and won the $1 million prize (Sill et al, 2009). In the eld of software engineering, applications of SG include predicting the numbers of remaining defects in black-box testing (Li et al, 2011), and malware detection in smartphones (Amamra et al, 2012). In a previous pilot study, we initially evaluated using SG for bug assignment with promising results (Jonsson et al, 2012).…”