Abstract :

Financial Fraud And Money Laundering Schemes Have Evolved Into Complex, Interconnected Networks That Traditional Tabular Analysis Methods Consistently Fail To Uncover. Standard Machine Learning Classifiers Evaluate Transactions As Isolated Data Rows, Rendering Them Blind To The Structural Dependencies Between Accounts That Define Modern Financial Crime. This Paper Proposes A Robust Fraud Detection Framework Using An Inductive Graph Neural Network (GNN), Specifically GraphSAGE (Graph Sample And Aggregate), Applied To The Elliptic Bitcoin Dataset. The Dataset Comprises 203,769 Transactions Represented As A Directed Graph Where Nodes Correspond To Wallet Addresses And Edges Represent Fund Flows. The Model Is Trained On A Strict Temporal Split—time Steps 1–34 For Training And 35–49 For Testing—to Simulate Real-world Concept Drift And Prevent Data Leakage. To Address Severe Class Imbalance (~2% Illicit Nodes), An Inverse-frequency Class-weighted Cross-entropy Loss Function Is Employed. The Proposed System Achieves An Illicit-class Recall Of 65.65% And F1-Score Of 0.507, Outperforming A Random Forest Baseline By Over 21 Percentage Points In Recall. A Streamlit Dashboard With Explainable AI (XAI) Reasoning And Interactive PyVis Topology Maps Enables Real-time Compliance Use

Published:

29-11-2024

Issue:

Vol. 24 No. 11 (2024)

Page Nos:

323 - 329

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite