Table 6 A summary of the performance of the proposed integrated schemes RF + SW-CNN (random forest with SW-CNN) and RF + SW-LSTM (random forest with SW-LSTM)
From: Use of subword tokenization for domain generation algorithm classification
| Â | Average precision | Average F1 | Average recall |
|---|---|---|---|
Proposed RF + SW-CNN | |||
Overall | 0.7893 | 0.7812 | 0.7895 |
Improvement over RF (Saeed et al. 2021) | 7.86% | 7.92% | 7.93% |
Improvement over SW-CNN | 4.00% | 6.59% | 5.52% |
Improvement over CNN-BiLSTM (Cucchiarelli et al. 2021) | 3.90% | 4.46% | 3.92% |
Random-looking DGAs | 0.7487 | 0.7377 | 0.7472 |
Improvement over RF (Saeed et al. 2021) | 0.44% | 0.23% | 0.43% |
Improvement over SW-CNN | 6.06% | 9.65% | 7.81% |
Improvement over CNN-BiLSTM (Cucchiarelli et al. 2021) | 4.15% | 5.00% | 4.23% |
Word-looking DGAs | 0.9300 | 0.9327 | 0.9391 |
Improvement over RF (Saeed et al. 2021) | 37.13% | 36.76% | 35.51% |
Improvement over SW-CNN | − 0.59% | − 0.40% | 0% |
Improvement over CNN-BiLSTM (Cucchiarelli et al. 2021) | 3.58% | 3.52% | 3.70% |
Proposed RF + SW-LSTM | |||
Overall | 0.7904 | 0.7833 | 0.7907 |
Improvement over RF (Saeed et al. 2021) | 8.01% | 8.21% | 8.05% |
Improvement over SW-LSTM | 4.44% | 6.71% | 5.60% |
Improvement over CNN-BiLSTM (Cucchiarelli et al. 2021) | 4.04% | 4.74% | 4.08% |
Random-looking DGAs | 0.7487 | 0.7382 | 0.7467 |
Improvement over RF (Saeed et al. 2021) | 0.44% | 0.30% | 0.36% |
Improvement over SW-LSTM | 6.49% | 9.67% | 7.89% |
Improvement over CNN-BiLSTM (Cucchiarelli et al. 2021) | 4.15% | 5.07% | 4.16% |
Word-looking DGAs | 0.9409 | 0.9409 | 0.9409 |
Improvement over RF (Saeed et al. 2021) | 38.73% | 37.96% | 35.77% |
Improvement over SW-LSTM | − 0.68% | − 0.29% | 0% |
Improvement over CNN-BiLSTM (Cucchiarelli et al. 2021) | 4.79% | 4.43% | 3.90% |