Date of Award

5-16-2025

Publication Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

Keywords

Graph Anomaly Detection; Graph level Anomaly Detection; Graph Neural Networks; Rayleigh Quotient

Supervisor

Ziad Kobti

Rights

info:eu-repo/semantics/openAccess

Abstract

Anomaly Detection (AD) is crucial across various domains, as it identifies irregularities or unusual patterns that, if quickly addressed, can prevent financial and data losses, protect health, and prevent disasters. Many systems such as social networks, communication systems, and biological networks are naturally represented as graphs with entities as nodes and interactions as edges. By analyzing these structures, we can uncover anomalies that are not apparent using traditional methods. However, current graph-based AD techniques face significant challenges, particularly suffering from low accuracy on larger datasets. As datasets grow larger, the complexity of the graphs increases. This complexity makes it more challenging for models to distinguish normal variations from true anomalies. Moreover, existing Graph Neural Network (GNN) algorithms focus primarily on spatial domain features while neglecting spectral properties. Furthermore, most existing algorithms focus on intra-graph properties (e.g., node and edge features), while overlooking the rich global inter-graph relationships, including graph similarity measures and cross-graph connectivity. To address these limitations, we propose a hybrid method called RQPool that integrates RQGNN-based intra-graph spectral properties with a variety of inter-graph spatial pooling strategies, such as sort, mean, Set2Set, GraphSAGE variants (mean and mean+max), SAGPooling, TopK pooling, and attention-based pooling, into a unified graph-level anomaly detection classifier. In empirical evaluations across multiple datasets, RQPool consistently achieves higher AUC and macro-F1 scores than methods based solely on spectral or spatial features, particularly excelling on large-scale graphs.

Download

Share

COinS