Back to search

North Dakota State University

Novel Techniques Using Graph Neural Networks (GNNS) for Anomaly Detection

Abstract

dc:description.abstract

This paper explores 2 new mechanisms that leverage graphs for anomaly detection. The novelty in approach one is to leverage the global attention capability of transformer architecture using a Graph Attention Network (GAT) with Chebyshev Laplacian for representation. This method leverages the GAT to learn attention weights for the graph features obtained through Chebyshev expansion of the Laplacian. This method focuses on capturing higher-order graph features with reduced computational complexity and utilizing attention mechanisms for improved feature relevance in detecting anomalies. The second approach leverages Fisher information to find anomalous graphs with ChebNet module for graph analysis. The ChebNet module allows for deep learning on graphs, capturing complex patterns and relationships that can help in detecting fraud more accurately. Using Fisher information improves model interpretability while ChebNet modules help leverage spectral properties.

Degree

thesis:*
Grantor dc:publisher
North Dakota State University
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Joseph, Priya

Rights

dc:rights
Statement dc:rights
  • NDSU policy 190.6.2

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/10365/33244

Chain of custody

source
Harvested from
North Dakota State University
Base URL
library.ndsu.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Joseph, Priya. Novel Techniques Using Graph Neural Networks (GNNS) for Anomaly Detection. North Dakota State University, 2023. https://hdl.handle.net/10365/33244