An Improved Method for Cross-Project Defect Prediction by Simplifying Training Data

Table 9

A comparison between our approach and two baseline methods for the data sets from PROMISE and AEEEM. The comparison is conducted based on the best prediction results of all the three methods in question.

Test set	Baseline1	Baseline2	Euclidean + Linear
Ant	0.785	0.803	1.3%	−1.0%	Baseline1 vs. TDSelector: −0.409
Xalan	0.657	0.675	10.7%	7.7%
Camel	0.595	0.624	0.5%	−4.2%
Ivy	0.789	0.802	4.7%	3.0%
Jedit	0.694	0.782	14.3%	1.4%
Lucene	0.608	0.701	−0.8%	−14.0%
Poi	0.691	0.789	3.3%	−9.5%
Synapse	0.740	0.748	2.3%	1.2%
Velocity	0.330	0.331	65.2%	64.7%	Baseline2 vs. TDSelector: −0.009
Xerces	0.714	0.753	8.5%	2.9%
Eclipse	0.706	0.744	10.2%	4.6%
Equinox	0.587	0.720	23.1%	0.3%
Lucene2	0.705	0.724	2.5%	−0.2%
Mylyn	0.631	0.646	9.3%	6.8%
Pde	0.678	0.737	10.4%	1.5%
Avg.	0.663	0.705	10.6%	4.3%