Link Prediction Methods for Complex Illicit Networks: A Systematic Review

Illicit trades, organized crime, trafficking, and terrorism remain major threats to global security, yet the networks that sustain these activities are difficult to study because available data are typically secondary, incomplete, and strategically con- cealed. In such settings, link prediction becomes important for inferring missing or unobserved relationships from network structure and associated attributes. This paper presents a systematic review focused on link prediction in illicit net- works. We synthesize the network representations, input features, models, and evaluation practices used across the existing literature, and critically examine how different link prediction methods have been adapted to the complex, covert, dynamic, and incomplete nature of illicit systems. The review shows that current studies span a wide range of illicit contexts and employ diverse methodological approaches, from similarity-based indices to graph neural networks and learning- based methods. However, the literature remains fragmented and is still dominated by static formulations, limited data settings, and evaluation strategies that do not fully reflect real-world investigative conditions. We identify major methodological and empirical gaps, including strong closed-world assumptions, and insufficient attention to robustness and uncertainty. The review concludes by outlining key directions for future research aimed at developing more realistic, interpretable, and operationally useful link prediction models for illicit-network analysis.