The effect of granularity level on software defect prediction

Application of defect predictors in software development helps the managers to allocate their resources such as time and effort more efficiently and cost effectively to test certain sections of the code. In this research, we have used Naïve Bayes Classifier (NBC) to construct our defect prediction framework. Our proposed framework uses the hierarchical structure information about the source code of the software product, to perform defect prediction at a functional method level and source file level. We have applied our model on SoftLAB and Eclipse datasets. We have measured the performance of our proposed model and applied cost benefit analysis. Our results reveal that source file level defect prediction improves the verification effort, while decreasing the defect prediction performance in all datasets.