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

Table 5

The best prediction results obtained by the CPDP approach based on TDSelector with Cosine similarity. NoD represents the baseline method; + denotes the growth rate of AUC value; the maximum AUC value of different normalization methods is underlined; each number shown in bold indicates that the corresponding AUC value rises by more than 10%.

Cosine similarity	Ant	Xalan	Camel	Ivy	Jedit	Lucene	Poi	Synapse	Velocity	Xerces	Eclipse	Equinox	Lucene2	Mylyn	Pde	Mean ± St.d
Linear
α	0.7	0.9	0.9	1.0	0.9	1.0	0.9	1.0	0.6	0.9	0.8	0.6	0.7	0.7	0.5		0.338
AUC	0.813	0.676	0.603	0.793	0.700	0.611	0.758	0.741	0.512	0.742	0.783	0.760	0.739	0.705	0.729	0.711 ± 0.081
+ (%)	6.3%	3.7%	1.9%	-	30.6%	-	3.0%	-	43.0%	0.3%	22.6%	39.4%	4.1%	5.9%	4.0%	9.0%
Logistic
α	0.7	0.5	0.7	1	0.7	0.6	0.6	0.6	0.5	0.5	0	0.4	0.7	0.5	0.5		0.351
AUC	0.802	0.674	0.595	0.793	0.665	0.621	0.759	0.765	0.579	0.745	0.773	0.738	0.712	0.707	0.740	0.711 ± 0.070
+ (%)	4.8%	3.4%	0.5%	-	24.1%	1.6%	3.1%	3.2%	61.7%	0.7%	21.0%	35.5%	0.3%	6.2%	5.6%	9.0%
Square root
α	0.7	0.7	0.6	0.6	0.7	0.6	0.7	0.9	0.5	1	0.4	0.6	0.6	0.6	0.6		0.249
AUC	0.799	0.654	0.596	0.807	0.735	0.626	0.746	0.762	0.500	0.740	0.774	0.560	0.722	0.700	0.738	0.697 ± 0.091
+ (%)	4.4%	0.3%	0.7%	1.8%	37.1%	2.5%	1.4%	2.8%	39.7%	-	21.0%	2.8%	1.7%	5.3%	5.3%	6.9%
Logarithmic
α	0.6	0.6	0.9	1.0	0.7	1.0	0.7	0.7	0.5	0.9	0.5	0.5	0.6	0.6	0.6		0.351
AUC	0.798	0.662	0.594	0.793	0.731	0.611	0.748	0.744	0.500	0.758	0.774	0.700	0.755	0.702	0.741	0.707 ± 0.083
+ (%)	4.3%	1.5%	0.3%	-	36.4%	-	1.6%	0.4%	39.7%	2.4%	21.2%	28.5%	6.3%	5.5%	5.8%	8.5%
Inverse cotangent
α	0.7	1.0	1.0	1.0	0.7	1.0	0.7	1.0	0.6	0.7	0	0.7	0.7	0.7	0.7		0.213
AUC	0.798	0.652	0.592	0.793	0.659	0.611	0.749	0.741	0.500	0.764	0.773	0.556	0.739	0.695	0.734	0.690 ± 0.092
+ (%)	4.3%	-	-	-	22.9%	-	1.8%	-	39.7%	3.2%	21.0%	2.1%	4.1%	4.4%	4.8%	5.9%
NoD	0.765	0.652	0.592	0.793	0.536	0.611	0.736	0.741	0.358	0.740	0.639	0.543	0.709	0.665	0.701	0.652 ± 0.113