Research Article
VulDistilBERT: A CPS Vulnerability Severity Prediction Method Based on Distillation Model
Table 8
Performance comparison between pretrained model and fine-tuned model.
| Models | Pretrained DistilBERT | Fine-tuned DistilBERT | Improvement (%) |
| Best epoch | 24 | 4 | 500 | Loss | 1.015 | 0.571 | 43.74 | Accuracy | 0.56 | 0.78 | 28.21 | Precision | 0.42 | 0.76 | 43.59 | Recall | 0.35 | 0.68 | 48.53 | F1 score | 0.34 | 0.70 | 51.43 |
|
|