Multi-stage latent graph learning for internet of things intrusion detection

Rajakumar Ponnumani, Anusuya Venkatasamy, Selvaraj Palanisamy, Padmanabhan P, Thenmozhi Elumalai, Rajakumar Perumal

Abstract


Rapid expansion of connected internet of things (IoT) infrastructures has led to increasingly heterogeneous network traffic, making timely identification of malicious activities more difficult. Conventional intrusion detection approaches frequently experience degraded performance because redundant traffic attributes obscure abnormal communication patterns and limit contextual interpretation of suspicious flows. To address these limitations, this study introduces a multi-stage latent graph learning (MS-LGL) framework that performs intrusion analysis through successive refinement of network traffic. The framework first compresses high-dimensional flow features into a compact latent space using a variational autoencoder (VAE). An isolation forest (IF) subsequently isolates suspicious traffic instances directly from the latent representations without requiring labeled attack data. Only these suspicious instances are converted into graph structures, allowing a graph attention network (GAT) to infer attack categories by exploiting structural dependencies among related communication flows. The proposed approach was validated using the CIC IoT-DIAD 2024 benchmark and assessed through standard intrusion detection metrics while being compared with representative state-of-the-art methods. The proposed framework attained 99.08% classification accuracy, a macro-average F1-score of 98.04%, and a macro-average receiver operating characteristic–area under the curve (ROC-AUC) of 0.986. These results indicate that progressive latent-space filtering combined with graph-based reasoning provides accurate and reliable multi-class intrusion detection for IoT networks.

Keywords


Graph attention network; Internet of things; Intrusion detection system; Isolation forest; Latent representation learning; Multi-class intrusion detection; Variational autoencoder

Full Text:

PDF


DOI: https://doi.org/10.11591/eei.v15i6.13931

Refbacks

  • There are currently no refbacks.


Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Bulletin of EEI Statistics

Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191 , e-ISSN: 2302-9285
This journal is published by the Institute of Advanced Engineering and Science (IAES) .