🛡️ TCH-Net — Cross-Domain IoT Botnet Detector

Multi-Branch Neural Network trained on the BRIDGE benchmark

TCH-Net achieves F1 = 0.8296 · ROC-AUC = 0.9380 across 5 structurally distinct IoT datasets, outperforming 12 baselines. Three parallel branches (Temporal, Statistical, Contextual) are fused via CB-GAF (Cross-Branch Gated Attention Fusion) — the first formally quantified generalisation baseline in heterogeneous IoT intrusion detection.

Select a dataset from the BRIDGE benchmark. Each example uses real flows sampled from the BRIDGE test split, with equal attack and benign rows mixed together.

Traffic Scenario

Override Dataset Source

Override Device Category

Window-level Results

Architecture: TCHNetV3 — 2.69M params · 6.43ms latency (Tesla T4) · Jetson-deployable
Benchmark: BRIDGE — 5 heterogeneous IoT datasets · 46 canonical features · 2.37M flows
Author: Ammar Bhilwarawala Citation: arXiv:2604.11324