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

Table 7

The best prediction results obtained by the CPDP approach based on TDSelector with Manhattan distance.

Manhattan distance	Ant	Xalan	Camel	Ivy	Jedit	Lucene	Poi	Synapse	Velocity	Xerces	Eclipse	Equinox	Lucene2	Mylyn	Pde	Mean ± St.d
Linear
α	0.8	0.9	0.9	1.0	0.9	0.9	1.0	1.0	0.8	1.0	0	0.8	0.9	1.0	1.0		0.187
AUC	0.804	0.753	0.599	0.816	0.689	0.626	0.695	0.748	0.500	0.749	0.773	0.633	0.692	0.695	0.668	0.696 ± 0.084
+ (%)	1.3%	7.0%	0.3%	-	7.3%	6.3%	-	-	7.8%	-	11.6%	19.0%	39.7%	-	-	5.6%
Logistic
α	0.7	0.7	0.8	0.8	0.8	0.7	0.7	0.9	0.6	0.7	0	0.9	0.9	1.0	1.0		0.249
AUC	0.799	0.760	0.607	0.830	0.674	0.621	0.735	0.794	0.520	0.756	0.773	0.680	0.559	0.695	0.668	0.705 ± 0.084
+ (%)	0.6%	8.0%	1.7%	1.7%	5.0%	5.4%	5.8%	6.1%	12.1%	0.9%	11.6%	27.9%	12.7%	-	-	6.9%
Square root
α	0.9	0.9	0.9	1.0	0.8	0.8	0.9	0.8	0.9	1.0	0	1.0	0	1.0	1.0		0.164
AUC	0.795	0.755	0.604	0.816	0.693	0.627	0.704	0.750	0.510	0.749	0.773	0.532	0.523	0.695	0.668	0.680 ± 0.1
+ (%)	0.1%	7.2%	1.2%	-	7.9%	6.5%	1.3%	0.3%	9.9%	-	11.6%	-	4.6%	-	-	3.1%
Logarithmic
α	1.0	0.9	0.9	1.0	0.9	1.0	1.0	0.8	0.9	0.9	0	1.0	0	1.0	1.0		0.116
AUC	0.794	0.755	0.603	0.816	0.664	0.589	0.695	0.763	0.524	0.756	0.773	0.532	0.523	0.695	0.668	0.677 ± 0.102
+ (%)	-	7.2%	1.0%	-	3.4%	-	-	2.0%	12.9%	0.9%	11.6%	-	4.6%	-	-	2.7%
Inverse cotangent
α	1.0	0.9	0.9	0.9	0.9	0.8	0.9	1.0	0.7	0.8	0	1.0	0	1.0	1.0		0.133
AUC	0.794	0.749	0.608	0.821	0.667	0.609	0.710	0.748	0.500	0.758	0.773	0.532	0.523	0.695	0.668	0.677 ± 0.103
+ (%)	-	6.4%	1.8%	0.6%	3.9%	3.4%	2.2%	-	7.8%	1.2%	11.6%	-	4.6%	-	-	2.7%
NoD (α = 1)	0.794	0.704	0.597	0.816	0.642	0.589	0.695	0.748	0.464	0.749	0.693	0.532	0.500	0.695	0.668	0.659 ± 0.105