International Journal of Scientific Research and Engineering Development

The growing amount and sophistication of financial. deals have rendered the process of detecting fraud as a challenge to. modern financial systems. Conventional machine learning techniques. do not capture transactions often and analyze them individually. relational patterns that can seek to denote coordinated fraudulent. activities. GNNs are effective. solution modeled as an interrelation of transactions, enabling learning feature-level as well as structural in- formation. Nevertheless, the use of GNNs on massive transactions. graphs is highly scalable because of the high challenges. computational and memory requirements. This paper proposes a graph-based framework of fraud detection that is scalable. SAGE architecture that makes use of neighbor sampling and mini- batch training to facilitate effective learning on huge transactions. networks. The approach presented is a representation of transactions as. graph nodes shares with captures relational dependencies. transaction attributes. IEEE- Experimental evaluation. The scalable GNN has been shown to have succeeded in detecting CIS fraud in dataset. Framework attains a ROC-AUC score of 0.723 and gets better. fraud recall versus full-graph GNN training and significantly lowering the computational needs. The results high- light the performance of scalable graph-based fraud. identify and prove that it can be successfully implemented. inmassive financial economies.

Dasari Veera Venkata Nooka Surya Saran, Korivi Dhilly Srikanth Reddy, Ridhi, Pabolu Syama Sahtya Sri Phani Pradeep, Bhukya Rohith,"Enhancing Fraud Detection using Scalable Graph Neural Networks GNN" International Journal of Scientific Research and Engineering Development, V9(1): Page(1872-1877) Jan-Feb 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.