An Automatic Source Code Vulnerability Detection Approach Based on KELM

Table 7

Effect of different symbolization levels on Bi-LSTM precision.

Dataset	Symbolization group	FPR (%)	TPR (%)	P (%)	F1 (%)
BE-ALL	F	2.9	86.2	91.2	88.6
	F + V	3.9	83.1	88.1	85.5
	F + T	3.2	84.6	90.4	87.4
	F + V + T	3.5	84.5	89.4	86.9
RM-ALL	F	2.8	92.2	94.1	93.1
	F + V	3.8	90.3	91.9	91.1
	F + T	2.6	93.5	94.5	94.0
	F + V + T	3.4	90.9	92.7	91.8
HY-ALL	F	3.2	87.8	92.0	89.8
	F + V	3.3	83.8	91.0	87.2
	F + T	3.2	85.6	91.7	88.5
	F + V + T	3.3	86.0	91.1	88.5