Answers are grounded in this paper and cite their sources.
Proposed framework improves anomaly detection in IoT by employing hierarchical autoencoders and density-based clustering.
Abstract Smart farming’s IoT-driven automation and real-time analytics enhance productivity but simultaneously expose heterogeneous, resource-limited devices to advanced cyber threats. Flow-based, unsupervised threat detection offers a privacy-preserving alternative to signature-based methods, yet key challenges remain: autoencoders may lose subtle anomaly cues during compression, reconstruction-error thresholds can fail under dynamic traffic conditions, and density-based clustering often degrades in noisy, high-dimensional spaces. To address these issues, we propose a fog-aware, two-stage hierarchical autoencoder (AE) with latent-space gating, followed by Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for unsupervised attack categorization. In Stage 1, a shallow AE performs progressive dimensionality reduction, producing a compact 21-dimensional latent representation that preserves nonlinear patterns while meeting fog-node constraints. In Stage 2, a deep AE applies reconstruction-error scoring to detect anomalies, inherently denoising and isolating malicious behaviors. Only high-error latent vectors are forwarded to DBSCAN, improving separation between threat types, mitigating noise, and avoiding predefined cluster counts. Evaluated on the CIC IoT-DIAD 2024 dataset, the framework achieves 98.99% accuracy, 0.9897 F1-score, an Adjusted Rand Index (ARI) of 0.895, and a Davies–Bouldin Index (DBI) of 0.019, surpassing state-of-the-art methods in detecting and categorizing DoS, scanning, botnet, and man-in-the-middle attacks without labeled data or fixed signatures.
No takes yet. Share an insight, caveat, or question.
Manikandan et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: