An Approach to Semantic and Structural Features Learning for Software Defect Prediction

Table 3

F-measure values of defect prediction built with four types of features.

Projects	SDP-base	SDP-S1 (△%)	SDP-S2 (△%)	SDP-S2S (△%)
Camel	0.2531	0.5044 (99.3%)	0.3081 (21.7%)	0.5049 (99.5%)
Lucene	0.5912	0.6397 (8.2%)	0.6873 (16.3%)	0.7564 (27.9%)
Poi	0.7525	0.7250 (−3.7%)	0.7892 (4.9%)	0.7340 (−2.5%)
Synapse	0.4244	0.4444 (4.7%)	0.5204 (22.6%)	0.4390 (3.4%)
Xalan	0.6165	0.6780 (10.0%)	0.6229 (1.0%)	0.6623 (7.4%)
Xerces	0.1856	0.2406 (29.6%)	0.2432 (31.0%)	0.2650 (42.8%)

△% represents the growth rate of performance relative to SDP-base.