Post-2024 FL/FTL IDS benchmarks (verbatim figures) versus FedGen–AI
| Study and year | Methodology | Accuracy (%) | macro-F1 | Notable features |
|---|---|---|---|---|
| Fine-Tuning FL-IDS for Transportation IoT (2025) (Akinie et al., 2025) | FedFT-3: server pre-train, edge fine-tune | 99.2 | n/r | –42% edge RAM, −75% training time |
| Efficient FTL for Smart-Farming (Praharaj et al., 2025) | EfficientNet-B0 head-only FedAvg | 96–97 | 0.97 | 25 kB/round uplink (low-rank updates) |
| Fed 2-Stage Transformer for CAN (2025) (Zhang et al., 2025b) | Self-sup. CAN-BERT, full fine-tune | 99.8 | 0.99 | 12 MB model, 1.4 MB/round |
| FedGen–AI (our proposal) | FTL + GenAI + blockchain | 92.0† | 0.8015† | Cross-domain (IoT Enterprise), poisoning-resilient |
| Study and year | Methodology | Accuracy (%) | macro-F1 | Notable features |
|---|---|---|---|---|
| FedFT-3: server pre-train, edge fine-tune | 99.2 | n/r | –42% edge RAM, −75% training time | |
| EfficientNet-B0 head-only FedAvg | 96–97 | 0.97 | 25 kB/round uplink (low-rank updates) | |
| Self-sup. CAN-BERT, full fine-tune | 99.8 | 0.99 | 12 | |
Note(s):†Best per-client score after 20 global rounds; second client: 86.26%/0.7989
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.