Abstract
Effective prediction of defect-prone software modules enables software developers to avoid the expensive costs in resources and efforts they might expense, and focus efficiently on quality assurance activities. Different classification methods have been applied previously to categorize a module in a system into two classes; defective or non-defective. Among the successful approaches, finite mixture modeling has been efficiently applied for solving this problem. This paper proposes the shifted-scaled Dirichlet model (SSDM) and evaluates its capability in predicting defect-prone software modules in the context of four NASA datasets. The results indicate that the prediction performance of SSDM is competitive to some previously used generative models.