Balanced Adversarial Tight Matching for Cross-Project Defect Prediction

Table 5

The results of the balanced accuracy comparison with nine methods.

Target project	LR	NNFilter	TCA	TCA+	DBN	DPCNN	TCNN	MANN	ADA	Ours
Ant	0.547	0.589	0.590	0.562	0.525	0.558	0.543	0.620	0.772	0.873
Camel	0.507	0.531	0.512	0.516	0.509	0.515	0.511	0.606	0.692	0.739
Forrest	0.516	0.484	0.454	0.399	0.469	0.485	0.449	0.561	0.445	0.403
Ivy	0.511	0.562	0.533	0.544	0.508	0.540	0.537	0.586	0.743	0.896
Log4j	0.481	0.486	0.507	0.501	0.503	0.506	0.536	0.461	0.440	0.380
Lucene	0.571	0.564	0.575	0.544	0.536	0.591	0.582	0.536	0.682	0.711
Poi	0.521	0.595	0.554	0.543	0.555	0.604	0.602	0.600	0.704	0.745
Synapse	0.524	0.598	0.584	0.620	0.531	0.574	0.553	0.640	0.753	0.818
Velocity	0.523	0.585	0.560	0.548	0.529	0.571	0.549	0.603	0.626	0.780
Xalan	0.438	0.439	0.481	0.493	0.488	0.514	0.521	0.504	0.651	0.796
Xerces	0.539	0.581	0.563	0.540	0.533	0.581	0.553	0.506	0.634	0.712
Average	0.516	0.547	0.538	0.528	0.517	0.549	0.540	0.566	0.649	0.714
W (T) (L)	9/0/2	9/0/2	9/0/2	10/0/1	9/0/2	9/0/2	9/0/2	9/0/2	9/0/2	—
Improvement	38.3%	30.6%	32.8%	35.2%	38.1%	30.1%	32.3%	26.2%	10.1%	—
p-Value	0.008	0.008	0.008	0.006	0.008	0.010	0.013	0.013	0.026	—