The rapid expansion of digital payments, e-commerce and connected systems has intensified the risk of cyber fraud, spanning phishing, account takeover, financial fraud and IoT/ICS manipulation. This systematic review synthesizes evidence from 105 peer-reviewed studies published between 2015 and 2025, identified through Google Scholar, Web of Science and Scopus, to examine detection and prevention techniques. Guided by PRISMA methodology, the review categorizes approaches into machine learning (ML), deep learning (DL), rule-based systems, hybrid frameworks and governance driven prevention strategies. Findings indicate that ML, DL-based detection dominates the literature, achieving high reported accuracy and recall but with limited real-world applicability due to reliance on outdated public benchmarks, severe class imbalance and scarce operational datasets. Prevention focused approaches, including privacy preserving learning, cyber threat intelligence pipelines and hybrid human AI frameworks, remain underexplored despite their potential to reduce fraud incidence and financial loss. Across domains, consistent gaps emerge in dataset transparency, reproducibility and the reporting of deployment readiness metrics such as latency, cost and interpretability. This review provides an integrated evidence map linking detection to prevention, highlights methodological and operational shortcomings and outlines priorities for developing scalable, transparent and adversary resilient fraud defenses.