Enhancing Elliptic Curve Cryptography with Graph Neural Networks and Ensemble Methods

Cryptocurrencies, most notably Bitcoin, have revolutionized the world of finance with the introduction of decentralized and borderless transactions with unimaginable transparency. The reward is, however, the gradually increasing illicit money laundering, ransomware, and fraud this innovation has attracted and thus created great problems for regulators and other financial institutions. Illicit transaction identification and classification remain very important in sustaining the integrity of such networks. Traditional fraud detection methods often fail to work properly owing to the inherent complexity in the data and anonymity of participants. This paper proposes a novel approach that combines the power of Graph Neural Networks with ensemble machine learning methods for an effective classification of Bitcoin transactions into illicit or licit. We conduct our work based on the transaction data extracted from the Elliptic dataset, a comprehensive dataset containing both licit and illicit Bitcoin transactions, to develop fraudulent activity detection in such an ever-evolving and highly decentralized financial system. By combining the structural learning capabilities of GNNs with the robustness of ensemble techniques like XGBoost and Multilayer Perceptron (MLP), the proposed model achieves superior performance in detecting fraudulent activities. Our results show that this hybrid model significantly outperforms individual models, yielding an F1 score of $\mathbf{0. 8 3 7}$ compared to $\mathbf{0. 7 9 8}$ for GNN alone.