Although the value of using static code attributes to learn defect predictor has been widely debated, there is no doubt that software defect predictions can effectively improve software quality and testing efficiency. Many data mining methods have already been introduced into defect predictions. We noted there have several versions of defect predictor based on Naive Bayes theory, and analyzed their difference estimation method and algorithm complexity. We found the best one which is Multi- variants Gauss Naive Bayes (MvGNB) by performing prediction performance evaluation, and we compared this model with decision tree learner J48. Experiment results on the benchmarking data sets of MDP made us believe that MvGNB would be useful for defect predictions.
展开▼
