Research Article
An Automatic Source Code Vulnerability Detection Approach Based on KELM
Table 4
Effect of different neural network models on vulnerability detection precision.
| Dataset | Pattern | FPR (%) | TPR (%) | P (%) | F1 (%) |
| BE-ALL | + B | 2.9 | 82.0 | 91.7 | 86.6 | d + B | 3.9 | 83.1 | 88.1 | 85.5 | d + E | 4.4 | 81.9 | 86.8 | 84.3 | d + Ada-E | 3.9 | 82.7 | 88.1 | 85.3 | d + KE | 1.8 | 78.7 | 93.8 | 85.6 |
| RM-ALL | + B | 2.8 | 95.3 | 94.6 | 95.0 | d + B | 3.8 | 90.3 | 91.9 | 91.1 | d + E | 3.9 | 92.4 | 94.9 | 92.1 | d + Ada-E | 2.8 | 83.8 | 93.4 | 88.3 | d + KE | 1.1 | 82.7 | 97.4 | 89.5 |
| HY-ALL | + B | 5.1 | 83.9 | 86.9 | 85.4 | d + B | 3.3 | 83.8 | 91.1 | 87.2 | d + E | 4.4 | 83.6 | 88.3 | 85.9 | d + Ada-E | 3.8 | 84.3 | 89.8 | 87.0 | d + KE | 1.9 | 81.0 | 94.3 | 87.1 |
|
|