Addressing Heterophily in Graph Anomaly Detection: A Perspective of Graph Spectrum

Gao, Yuan; Wang, Xiang; He, Xiangnan; Liu, Zhenguang; Feng, Huamin; Zhang, Yongdong

doi:10.1145/3543507.3583268

Cited by 31 publications

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The proliferation of IoT devices has experienced a significant upswing in recent years [30], resulting in a substantial upsurge in the volume of data produced by these networked devices. The analysis of this extensive dataset is of the utmost importance for anomaly detection [31,32], as it facilitates the identification of atypical patterns or behaviors. Therefore, this literature review provides an overview of existing research that primarily focuses on active learning-based algorithms for anomaly detection in IoT systems.…”

Section: Literature Reviewmentioning

confidence: 99%

Anomaly Detection for IOT Systems Using Active Learning

Zakariah,

Almazyad

2023

Applied Sciences

View full text Add to dashboard Cite

The prevalence of Internet of Things (IoT) technologies is on the rise, making the identification of anomalies in IoT systems crucial for ensuring their security and reliability. However, many existing approaches rely on static classifiers and immutable datasets, limiting their effectiveness. In this paper, we have utilized the UNSW-NB15 dataset, which contains 45 variables including multi- and binary-target variables, to determine the most relevant properties for detecting abnormalities in IoT systems. To address this issue, our research has investigated the use of active learning-based algorithms for anomaly detection in IoT systems. Active learning is a powerful technique that improves precision and productivity by eliminating the need for labeling and adapting to dynamic IoT environments. Additionally, our study has combined feature engineering methods, active learning approaches, and a random forest classifier to construct a resilient anomaly detection model for IoT devices. The proposed model has outperformed several state-of-the-art techniques, achieving an impressive accuracy rate of 99.7%. By implementing a rigorous sampling procedure and leveraging the collaborative nature of the random forest technique, our model has demonstrated a notable level of precision with a weighted average accuracy of 0.995. The findings of the study offered empirical evidence, supporting the efficacy of our active learning methodology in identifying abnormalities in IoT systems. Moreover, our study provides valuable insights and recommendations for future research and development activities in this field. Overall, this research contributes to the advancement of anomaly detection techniques in IoT systems, further enhancing their security and reliability.

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Anomaly Detection for IOT Systems Using Active Learning

Zakariah,

Almazyad

2023

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…According to recent researches, scholars found the spectral 'right-shift' phenomenon (Tang et al 2022) on both synthetic and real datasets, proving that the higher the anomaly ratio, the more high-frequency components on the graph spectrum. In addition, researchers also demonstrate that, when using the ground truth labels as signals, the proportion of heterophily edges is positively correlated with the spectral energy distribution (Gao et al 2023a).…”

Section: Introductionmentioning

confidence: 99%

Revisiting Graph-Based Fraud Detection in Sight of Heterophily and Spectrum

Xu,

Wang,

et al. 2024

AAAI

View full text Add to dashboard Cite

Graph-based fraud detection (GFD) can be regarded as a challenging semi-supervised node binary classification task. In recent years, Graph Neural Networks (GNN) have been widely applied to GFD, characterizing the anomalous possibility of a node by aggregating neighbor information. However, fraud graphs are inherently heterophilic, thus most of GNNs perform poorly due to their assumption of homophily. In addition, due to the existence of heterophily and class imbalance problem, the existing models do not fully utilize the precious node label information. To address the above issues, this paper proposes a semi-supervised GNN-based fraud detector SEC-GFD. This detector includes a hybrid filtering module and a local environmental constraint module, the two modules are utilized to solve heterophily and label utilization problem respectively. The first module starts from the perspective of the spectral domain, and solves the heterophily problem to a certain extent. Specifically, it divides the spectrum into various mixed-frequency bands based on the correlation between spectrum energy distribution and heterophily. Then in order to make full use of the node label information, a local environmental constraint module is adaptively designed. The comprehensive experimental results on four real-world fraud detection datasets denote that SEC-GFD outperforms other competitive graph-based fraud detectors. We release our code at https://github.com/Sunxkissed/SEC-GFD.

show abstract