Journal of Information & Systems Management Hsing-Cheng Liu, Yao-Liang Chung 16 2 2026 https://doi.org/10.6025/jism/2026/16/2/37-55 https://www.dline.info/jism/fulltext/v16n2/jismv16n2_1.pdf The rapid expansion of e-commerce has intensified cybersecurity threats, particularly counterfeit product listings and fraudulent transactions, which severely undermine consumer trust and marketplace integrity. Traditional rule based monitoring systems increasingly struggle to detect these sophisticated, evolving fraud patterns. This study develops and evaluates a dual layer predictive analytics framework that leverages supervised machine learning to enhance counterfeit detection and fraud governance in digital marketplaces. Utilizing a Random Forest classifier, the research analyzes two complementary datasets: static productlevel metadata comprising seller credentials and listing characteristics, and dynamic transaction-level behavioral indicators capturing purchasing anomalies and device signatures. Following rigorous preprocessing, categorical encoding, and an 80/20 stratified train-test split, the models were benchmarked using accuracy, precision, recall, F1 Score, and ROC-AUC. The product-level classifier achieved perfect class separability, demonstrating the deterministic power of structural listing indicators. Conversely, the transaction-level model attained 99.33% accuracy, 97.99% precision, and a 0.9999 ROC-AUC, effectively capturing probabilistic behavioral fraud signatures with minimal false negatives. These findings confirm that integrating static asset metadata with dynamic transactional analytics significantly improves predictive robustness. The study concludes by proposing a unified, AI-driven marketplace governance framework capable of real-time surveillance, automated threat mitigation, and scalable regulatory compliance, ultimately advancing the deployment of transparent artificial intelligence solutions across global digital commerce networks.