Reviewer 3,Q9: Dataset quality

[Suvodeep90]

Furthermore, the dataset looks quite strange to me. Looking at the defect distributions (left-most boxplot in Figure 4), I do see that 75% of the projects (the last quartile) have more than 60% of defective modules (classes/methods?). Some have a defective percentage above 80%. Do we need machine learning to predict defects in projects where almost all modules are defectives? In these projects, a simple, constant classifier would work the best. The results also show this for ZeroR in Section 5.