Enhancing Anomaly Detection in Attributed Networks Using Proximity Preservation and Advanced Embedding Techniques

The increasing relevance of anomaly detection in attributed networks is gaining traction in fields such as cybersecurity, finance, and healthcare. However, large-scale attributed networks often exhibit noisy and inconsistent node properties, which negatively affect anomaly detection accuracy and dis...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access Vol. 13; pp. 42777 - 42796
Main Authors: Khan, Wasim, Ishrat, Mohammad, Nadeem Ahmed, Mohammad, Abidin, Shafiqul, Husain, Mohammad, Izhar, Mohd, Zamani, Abu Taha, Rashid Hussain, Mohammad, Ali, Arshad
Format: Journal Article
Language:English
Published: IEEE 2025
Subjects:
Anomaly detection
generational adversarial network
Generative adversarial networks
Laplace equations
Laplacian sharpening
Laplacian smoothing
Medical services
Noise
Noise measurement
proximity preservation
Self-supervised learning
Smoothing methods
Topology
Training
ISSN:2169-3536, 2169-3536
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The increasing relevance of anomaly detection in attributed networks is gaining traction in fields such as cybersecurity, finance, and healthcare. However, large-scale attributed networks often exhibit noisy and inconsistent node properties, which negatively affect anomaly detection accuracy and disrupt the network's structure. A key challenge is maintaining the integrity of both network and node feature structures during the embedding process. To address this, we propose a novel approach that combines a Graph Convolution Auto encoder (GCAE) with self-supervised learning, proximity preservation, and adversarial training using Generative Adversarial Networks (GAN). First, Laplacian smoothing is applied to reduce noise in node properties, followed by Laplacian sharpening to highlight important features. These enhanced features are then fed into the GCAE, which encodes node attributes into a latent space using graph convolutional layers. Self-supervised tasks like attribute masking and edge prediction further enhance the GCAE's ability to capture the graph's structure. Additionally, proximity preservation ensures that the latent space reflects both first order and high-order proximity. The inclusion of GAN refines the embeddings, aligning them closer to the true distribution of the graph data. This method effectively preserves both node features and network structure, making the embedding robust and distinguishable. Empirical evaluations on four real-world datasets demonstrate that our approach surpasses state-of-the-art methods, setting a new benchmark for anomaly detection in attributed networks. Our framework has significant potential to advance both research and practical applications in anomaly detection.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2025.3544260