Asian Journal of Research in Computer Science

This study explores the role of artificial intelligence (AI)-driven cybersecurity models in mitigating fraud, smart contract vulnerabilities, and regulatory challenges in cryptocurrency platforms. Utilizing datasets such as the Elliptic Bitcoin Dataset, SolidiFI-Benchmark, CryptoScamDB, and CipherTrace AML Reports, this research employs Logistic Regression, Random Forest, and Reinforcement Learning (RL) for fraud detection and anomaly identification. The AI-based security model demonstrates a 5.2% increase in fraud detection accuracy over traditional rule-based methods while reducing false positives by 19.3%. However, the model exhibits a false negative rate of 98.9%, indicating challenges in fully capturing sophisticated fraud techniques. Regression analysis shows a strong inverse correlation (R² = 0.927) between AI adoption and fraud cases, where each 1% increase in AI adoption corresponds to a reduction of approximately 37 fraud cases.In real-world applicability, the proposed AI-driven models enhance scalability and real-time threat detection but require substantial computational resources, particularly for deep learning and RL-based techniques. Computational efficiency is optimized through federated learning and quantum-resistant AI security, ensuring robust yet privacy-preserving fraud detection. Despite its advantages, challenges such as adversarial AI attacks, regulatory inconsistencies, and scalability under high transaction loads persist.The study recommends self-supervised learning for fraud detection, improving interpretability in deep learning models, and developing AI-driven compliance frameworks to address ethical concerns. By integrating Machine Learning (ML), Deep Learning (DL), and Reinforcement Learning (RL), this study provides a novel approach to securing cryptocurrency transactions, offering actionable insights for researchers, financial institutions, and policymakers.