Reviewer 3,Q12: Question regarding ZeroR learner

[Suvodeep90]

Section 5 contains the following recommendation “As to ZeroR, we cannot recommend that approach. While ZeroR makes few mistakes (low ifas and low pfs), it scores badly on other measures (very low recalls and popt(20).” This argument does not convince me as ZeroR gets the best results for two metrics (as far as I understand). Please elaborate more on why we cannot use ZeroR. Besides, ZeroR may work relatively well because of projects with a very large number of defects. Perhaps, some in-depth analysis of the correlation between defect distributions and performances of the models would be very useful.