A Deep Learning-Based Multi-Layer Anomaly Detection Framework for Mitigating Emerging IoT Cyber Threats
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The expanding adoption of Internet of Things (IoT) technologies across critical and consumer infrastructures has introduced complex security challenges arising from heterogeneous devices, massive traffic volumes, and continuously evolving cyber threats. Conventional intrusion detection systems and static machine learning approaches are often inadequate for identifying sophisticated and previously unseen attacks due to limited feature representation and poor adaptability to dynamic traffic behaviors. To address these challenges, this work proposes a deep learning-driven multi-layer anomaly detection framework that distributes intelligence across device, edge, and cloud layers to achieve both efficient and accurate threat mitigation in IoT environments. Lightweight models deployed at the device layer perform rapid preliminary screening, while hybrid CNN-LSTM/GRU architectures at the edge layer capture spatial and temporal traffic dependencies for contextual analysis. For complex and ambiguous attack patterns, a high-capacity ConvLSTM model augmented with an attention mechanism operates at the cloud layer to extract long-term temporal correlations and contextual relevance. A weighted fusion strategy integrates anomaly scores from all layers, enabling refined decision-making and reducing false alarms. The framework was extensively evaluated using contemporary IoT security datasets, including TON_IoT, IoT-23, and N-BaIoT. Experimental results demonstrate that the proposed multi-layer fusion model achieves an overall detection accuracy of 99.1%, with precision, recall, and$\mathbf{F 1}$-score values of 98.8%, 98.5%, and 98.6%, respectively, outperforming traditional machine learning classifiers and single-layer deep learning models by a margin of$\mathbf{4}-\mathbf{1 2} \boldsymbol{\%}$. Furthermore, the hierarchical design significantly reduces false positive rates and computational overhead, making the framework suitable for real-time and large-scale IoT deployments. These findings highlight the effectiveness of distributed spatial-temporal learning and cross-layer fusion in strengthening the resilience of IoT ecosystems against emerging and zero-day cyber threats.
Fields of Study & Subjects
Network Security and Intrusion Detection
Internet Traffic Analysis and Secure E-voting
Internet of Things and AI
Publication details
- Journal / source
- International Conference on Intelligent Data Communication Technologies and Internet of Things (IDCIoT)
- Pages
- 412–417
- Publication date
- 4 February 2026
- Language
- EN