VulMPFF: A Vulnerability Detection Method for Fusing Code Features in Multiple Perspectives

Table 5

Results of multifeature fusion ablation experiments (%), I_rem indicates removal of a IRC.

	Big-Vulsub				Big-Vulfull				Devign				Reveal
	Acc	Pre	Rec	F1	Acc	Pre	Rec	F1	Acc	Pre	Rec	F1	Acc	Pre	Rec	F1
SEQ	65.62	59.44	84.71	69.86	90.78	34.52	14.44	20.36	51.83	46.49	60.75	52.67	81.26	42.71	25.68	32.07
CPGrem	70.57	58.42	87.34	70.01	89.72	28.38	17.85	21.95	53.26	48.83	62.15	54.69	83.48	41.44	30.67	35.25
SASTrem	93.22	92.28	95.82	94.02	91.14	29.17	40.34	33.86	63.28	56.31	81.02	66.44	93.06	43.89	56.15	49.27
ASTrem	91.11	91.15	92.24	91.69	90.03	21.62	32.53	25.98	61.63	52.24	76.22	61.96	88.03	33.28	53.57	41.05
CFGrem	92.37	92.35	93.13	92.74	90.35	27.76	30.28	28.97	61.34	52.19	79.53	63.06	90.22	35.46	54.68	44.72
DDGrem	93.21	92.16	95.65	93.87	91.2	23.35	35.08	28.04	62.34	52.18	80.8	63.5	91.37	38.98	56.27	46.06
CDGrem	94.24	92.38	95.5	93.91	92.22	23.33	34.99	27.99	62.39	52.26	80.91	63.41	92.35	38.9	56.33	46.02
SEQ + CPG	93.39	92.62	95.69	94.13	92.84	30.17	40.68	34.65	63.27	56.21	81.45	66.52	93.04	44.83	56.09	49.83
VulMPFF	95.64	92.34	96.1	94.18	94.26	30.69	42.45	35.62	64.13	56.23	84.83	67.63	93.29	45.92	56.28	50.57

The bolded values in the table indicate the best result values obtained.