Self-Organizing Maps-Assisted Variational Autoencoder for Unsupervised Network Anomaly Detection DOI Open Access

Hailong Huang,

Jiahong Yang,

Hang Zeng

et al.

Symmetry, Journal Year: 2025, Volume and Issue: 17(4), P. 520 - 520

Published: March 30, 2025

In network intrusion detection systems (NIDS), conventional supervised learning approaches remain constrained by their reliance on labor-intensive labeled datasets, especially in evolving ecosystems. Although unsupervised offers a viable alternative, current methodologies frequently face challenges managing high-dimensional feature spaces and achieving optimal performance. To overcome these limitations, this study proposes self-organizing maps-assisted variational autoencoder (SOVAE) framework. The SOVAE architecture employs correlation graphs combined with the Louvain community algorithm to conduct selection. processed data—characterized reduced dimensionality clustered structure—is subsequently projected through maps generate cluster-based labels. These labels are further incorporated into symmetric encoding-decoding reconstruction process of VAE enhance data quality. Anomaly is implemented quantitative assessment discrepancies SOM deviations. Experimental results show that achieves F1 scores 0.983 (±0.005) UNSW-NB15 0.875 (±0.008) CICIDS2017, outperforming mainstream baselines.

Language: Английский

Self-Organizing Maps-Assisted Variational Autoencoder for Unsupervised Network Anomaly Detection DOI Open Access

Hailong Huang,

Jiahong Yang,

Hang Zeng

et al.

Symmetry, Journal Year: 2025, Volume and Issue: 17(4), P. 520 - 520

Published: March 30, 2025

In network intrusion detection systems (NIDS), conventional supervised learning approaches remain constrained by their reliance on labor-intensive labeled datasets, especially in evolving ecosystems. Although unsupervised offers a viable alternative, current methodologies frequently face challenges managing high-dimensional feature spaces and achieving optimal performance. To overcome these limitations, this study proposes self-organizing maps-assisted variational autoencoder (SOVAE) framework. The SOVAE architecture employs correlation graphs combined with the Louvain community algorithm to conduct selection. processed data—characterized reduced dimensionality clustered structure—is subsequently projected through maps generate cluster-based labels. These labels are further incorporated into symmetric encoding-decoding reconstruction process of VAE enhance data quality. Anomaly is implemented quantitative assessment discrepancies SOM deviations. Experimental results show that achieves F1 scores 0.983 (±0.005) UNSW-NB15 0.875 (±0.008) CICIDS2017, outperforming mainstream baselines.

Language: Английский

Citations

0