Answers are grounded in this paper and cite their sources.
Evaluates detection performance in IIoT using CNN-LSTM, Transformer, and VAE models, indicating novel strategies for cybersecurity.
The proliferation of Internet of Things (IoT) devices in industrial automation, smart manufacturing, and critical infrastructure — collectively termed Industrial IoT (IIoT) — has created an expansive and heterogeneous attack surface that traditional signature-based intrusion detection systems (IDS) are ill-equipped to defend. Machine-to-machine communication protocols (MQTT, OPC-UA, Modbus TCP), legacy supervisory control and data acquisition (SCADA) systems, and resource-constrained sensor nodes operating in deterministic real-time environments present unique cybersecurity challenges that demand anomaly-based detection approaches capable of identifying zero-day attacks and novel attack variants from network traffic patterns alone, without prior knowledge of attack signatures. This paper presents a systematic comparative evaluation of three deep learning architectures — a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model, a multi-head self-attention Transformer encoder, and a Variational Autoencoder (VAE) — for network intrusion detection on the CIC-IDS-2018 benchmark dataset (16 million traffic flows, 14 attack categories). After class-imbalance correction via Synthetic Minority Oversampling Technique (SMOTE), the CNN-LSTM model achieves macro-average F1-score of 0.9724, outperforming the Transformer (0.9681) and VAE (0.9512) on the held-out test set. Critically, the CNN-LSTM demonstrates superior detection of low-volume attacks (Infiltration: F1 = 0.912) that pose the greatest operational risk in IIoT environments. Inference latency analysis confirms that the CNN-LSTM and Transformer models satisfy real-time detection requirements (latency < 2 ms per flow on NVIDIA RTX 3060) while the VAE's reconstruction-based detection introduces unacceptable latency for real-time deployment. The paper concludes with a deployment architecture recommendation for edge-cloud hierarchical IDS in IIoT environments.
No takes yet. Share an insight, caveat, or question.
Swapnil Deshmukh Anupam Borthakur (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: