A Software Defect Prediction Approach Based on Hybrid Feature Dimensionality Reduction

Table 5

The statistical test results of prediction performance of the HFDRA method and other comparison methods under SVM classifier.

Measure		Methods
		ALL	IG	MIC	KPCA	CFS	FCBF	ReliefF	HFDRA
AUC	AVE	0.811	0.807	0.808	0.836	0.727	0.654	0.805	0.835
	W/D/L	7/1/14	9/0/13	7/0/15	9/1/12	3/0/19	3/0/19	8/1/13	–
	p-value	0.434	0.387	0.392	0.950	0.004	0.000	0.335	–
	Cliff’s	0.188	0.194	0.194	0.043	0.546	0.822	0.234	–
F1	AVE	0.803	0.795	0.794	0.831	0.666	0.603	0.795	0.842
	W/D/L	5/1/16	7/0/15	6/0/16	7/0/15	3/0/19	2/0/20	5/1/16	–
	p-value	0.234	0.175	0.161	0.726	0.001	0.000	0.149	–
	Cliff’s	0.285	0.298	0.326	0.143	0.651	0.872	0.364	–
MCC	AVE	0.624	0.616	0.616	0.671	0.480	0.327	0.608	0.673
	W/D/L	6/1/15	8/0/14	7/0/15	10/1/11	3/0/19	3/0/19	8/1/13	–
	p-value	0.429	0.367	0.355	0.976	0.002	0.000	0.296	–
	Cliff’s	0.219	0.198	0.186	0.064	0.558	0.802	0.267	–