Misbehavior Detection on VeReMi
99.99AccuracyDeep CNN (IncResNet)
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Deep CNN (IncResNet)Platoon=No, Architecture=Deep CNN (IncResNet), Deploy=Appropriate HW, Inference Time [ms]=5-7, Attacks=DoS, spoofing2025.12 | 99.99 | 99 | 99 | — | — | — | — | — | — | — | — | — | |
| DNN with DBNPlatoon=No, Architecture=DNN with DBN, Deploy=Appropriate HW, Inference Time [ms]=2-5, Attacks=Pkt. inj., manip.2025.12 | 99.7 | 99 | 99 | — | — | — | — | — | — | — | — | — | |
| CNN + Attn GRUPlatoon=No, Architecture=CNN + Attn GRU, Attacks=Mixed intrusion2025.12 | 99 | — | — | — | — | — | — | 10.8 | — | — | — | — | |
| LSTM/CNN-LSTMPlatoon=No, Architecture=3/4/5x LSTM, CNN-LSTM, Deploy=Not appropriate HW, Attacks=Pos./spd. falsif., Controller=N/A2025.12 | 98 | 97 | 99 | — | — | — | — | — | — | — | — | — | |
| CNN-LSTM, GRU, AEPlatoon=No, Architecture=CNN-LSTM, GRU, AE, Attacks=Pos./spd. falsif.2025.12 | 98 | 97 | 96 | — | — | — | — | — | — | — | — | — | |
| CNN-LSTM + AttnPlatoon=No, Architecture=CNN-LSTM + Attn, Attacks=Multi-attack CAN2025.12 | 97.8 | — | — | 0.95 | — | — | — | — | 2.16 | — | — | — | |
| CNN-LSTM SVMPlatoon=No, Architecture=CNN-LSTM SVM, Attacks=Pos. falsif.2025.12 | 95.4 | — | — | 96.1 | — | — | — | — | — | — | — | — | |
| FedAvg, FedProxPlatoon=No, Architecture=FedAvg, FedProx, Attacks=19 VeReMi types2025.12 | 93 | 92 | 91 | 92 | — | — | — | — | — | — | — | — | |
| FedML (SVM, LSTM)Platoon=No, Architecture=FedML (SVM, LSTM), Attacks=Pos. falsif.2025.12 | 92 | 93 | 90 | — | — | — | — | — | — | — | — | — | |
| Multi-task LSTMPlatoon=No, Architecture=Multi-task LSTM, Deploy=Appropriate HW, Inference Time [ms]=0.61-144, Attacks=Multi-dim. anom.2025.12 | 90 | 89 | 88 | — | — | — | — | — | — | — | — | — | |
| LSTM AutoencoderPlatoon=No, Architecture=LSTM Autoencoder, Deploy=Appropriate HW, Inference Time [ms]=650, Attacks=Replay, manip.2025.12 | — | — | — | 0.95 | 0.97 | — | — | — | — | — | — | — |