Answers are grounded in this paper and cite their sources.
Evaluates a Bi-LSTM model for real-time intrusion detection in IoT networks, suggesting enhanced security management.
The exponential proliferation of Internet of Things (IoT) devices in academic and enterprise environments — with an estimated 18.8 billion connected devices globally as of 2024, projected to reach 40.6 billion by 2030 — has dramatically expanded the attack surface for network intrusions, malware propagation, and distributed denial-of-service (DDoS) campaigns. Conventional signature-based intrusion detection systems (IDS) are inadequate for IoT network environments characterised by heterogeneous device profiles, constrained computational resources, high packet rates, and the emergence of novel zero-day attack vectors that lack pre-defined signature patterns. Machine learning-based anomaly detection offers a path beyond signature limitations, but the temporal dependency structure of network traffic sequences — where the malicious intent of a packet may be discernible only in the context of preceding and succeeding traffic patterns — motivates the application of recurrent neural architectures capable of capturing bidirectional temporal context. This study proposes and evaluates a Bidirectional Long Short-Term Memory network augmented with a multi-head scaled dot-product attention mechanism (Bi-LSTM+Att) for intrusion detection in IoT-enabled smart campus network environments. The proposed architecture is trained and evaluated on the NSL-KDD benchmark dataset (125,973 training samples, 22,544 test samples across five traffic classes: Normal, DoS, Probe, R2L, and U2R) and validated on a custom campus IoT dataset collected from the smart building infrastructure of Chandigarh Engineering College comprising 48 heterogeneous IoT devices over a 60-day monitoring period. The proposed Bi-LSTM+Att model achieves 94.8% overall accuracy, 93.6% precision, 94.2% recall, and 93.9% F1-score on NSL-KDD — surpassing baseline LSTM (90.1%), SVM (87.3%), Random Forest (85.6%), and Naive Bayes (78.4%) under identical training and evaluation protocols. The attention mechanism provides interpretable feature attribution maps confirming that packet duration, source byte count, and protocol type are the dominant discriminative features for attack classification. Inference latency of 34 ms at 10,000 packets/second on a mid-range GPU confirms viability for real-time campus network deployment. These findings establish Bi-LSTM+Att as a practical, deployable IDS architecture for resource-aware IoT security management in Indian academic institution networks.
No takes yet. Share an insight, caveat, or question.
Nidhi Sharma Harpreet Singh Bhatia (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: