DB Online Gambling Detection System: A SLR

The rapid expansion of online gambling has produced a complex, multimodal, and cross-jurisdictional digital ecosystem that increasingly challenges cybersecurity, regulatory governance, and harm-prevention efforts. This study conducts a Systematic Literature Review (SLR) of 30 primary studies published between 2021 and 2025, synthesized using the PICOC framework to map detection architectures, data modalities, evaluation practices, and evasion behaviors. The evidence reveals a clear methodological progression: early rule-based and classical Machine Learning (ML) pipelines (TF\u2013IDF + Random Forest and SVM) have been superseded by Deep Learning (DL) architectures, particularly Transformer-based models (IndoBERT) for Indonesian-language promotion detection and hybrid multimodal systems that fuse screenshot features, OCR text, HTML structures, and network or blockchain signals. These multimodal frameworks consistently achieve state-of-the-art performance (F1-scores above 0.99). Yet, their real-world applicability remains constrained by adversarial camouflage (obfuscated URLs, visual overlays, injected pages), anti-bot and redirect-based acquisition barriers, scarce multilingual datasets, and pronounced temporal and jurisdictional drift. The review also highlights vigorous regional activity in Indonesia, notably in university and government web defacement cases involving hidden gambling pages and \slot backdoor\ injection patterns. By consolidating heterogeneous evidence into a unified taxonomy, this SLR identifies six persistent research gaps: limited multimodal and multilingual benchmarks, under-evaluated adversarial robustness, lack of continual and cross-jurisdiction learning, weak integration with operational cybersecurity workflows, insufficient explainability in high-stakes settings, and inadequate coverage of Web3 gambling ecosystems. The study concludes with a forward agenda emphasizing robustness audits, redirect- and anti-bot\u2013aware acquisition pipelines, explainable multimodal fusion, and lightweight, deployment-ready detection models suitable for edge and cloud environments.