Applying Generative AI for Fraud and Cybercrime Prevention in Thailand

The rise of fraudulent websites requires advanced real-time detection methods to protect individuals and organizations. This paper proposes a hybrid framework combining supervised and unsupervised learning to identify and classify fraudulent websites with high accuracy. The supervised model detects known malicious patterns using labeled data, while the unsupervised model identifies anomalies to flag emerging threats. Together, they ensure comprehensive detection, achieving 97.9% accuracy with a low latency of 35 milliseconds per website. The system includes a risk-scoring mechanism based on URL anomalies, SSL/TLS validity, and behavioral patterns, categorizing websites into Safe, Moderate, and High-risk levels. Experimental results show the framework outperforms traditional blacklist systems and standalone models, highlighting its potential for real-time fraud detection and future advancements in cyber threat mitigation.