Year: 2024
Type: article
Abstract: Online fraud detection is a critical aspect of modern cybersecurity, as fraudulent activities in financial transactions can lead to significant economic losses, reputational damage, and erosion of consumer trust. Failure to detect and mitigate fraud in real-time allows cybercriminals to exploit vulnerabilities, resulting in increased chargebacks, legal liabilities, and financial instability. Traditional fraud detection methods often struggle to keep pace with evolving fraud patterns, necessitating more advanced approaches. This research presents a graph-based fraud detection framework, demonstrating the superiority of GraphSAGE in identifying fraudulent transactions GraphSAGE excels by leveraging structural information and transaction relationships, capturing intricate fraud patterns that traditional machine learning models fail to detect. Unlike conventional models that analyze transactions in isolation, GraphSAGE constructs a transaction network, allowing it to identify suspicious behaviors based on contextual and relational features. By aggregating information from neighboring transactions through graph convolution techniques, it enhances fraud detection capabilities, especially in complex, high-volume financial ecosystems. The results demonstrate its superiority, achieving the highest accuracy (97.50%) and recall (96.79%), significantly outperforming traditional models such as XGBoost (94.47%) and LightGBM (94.44%). The findings highlight GraphSAGE as the most effective solution for online fraud detection, providing a scalable and robust approach to combating fraudulent activities in financial systems.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4411332332