[Paper Review] The web of federal crimes in Brazil: topology, weaknesses, and control

Bruno Requião da Cunha, Sebastián Gonçalves|arXiv (Cornell University)|Jun 9, 2017

Crime, Illicit Activities, and Governance18 citations

TL;DR

This study analyzes a large-scale, real-world network of Brazilian federal crimes involving 23,666 individuals and 35,913 relationships. Using network science, it reveals the network's high modularity (Q=0.96), extreme vulnerability to module-based attacks (disruption with only 2% edge/node removal), and mathematical controllability via 21% of driver nodes, offering a proactive strategy for law enforcement to dismantle criminal systems more efficiently than traditional methods.

ABSTRACT

Law enforcement and intelligence agencies worldwide struggle to find effective ways to fight and control organized crime. However, illegal networks operate outside the law and much of the data collected is classified. Therefore, little is known about criminal networks structure, topological weaknesses, and control. In this contribution we present a unique criminal network of federal crimes in Brazil. We study its structure, its response to different attack strategies, and its controllability. Surprisingly, the network composed of multiple crimes of federal jurisdiction has a giant component, enclosing more than a half of all its edges. This component shows some typical social network characteristics, such as small-worldness and high clustering coefficient, however it is much "darker" than common social networks, having low levels of edge density and network efficiency. On the other side, it has a very high modularity value, $Q=0.96$. Comparing multiple attack strategies, we show that it is possible to disrupt the giant component of the network by removing only $2\%$ of its edges or nodes, according to a module-based prescription, precisely due to its high modularity. Finally, we show that the component is controllable, in the sense of the exact network control theory, by getting access to $20\%$ of the driver nodes.

Motivation & Objective

• To understand the topological structure of a large-scale, real-world network of Brazilian federal crimes.
• To evaluate the robustness of this criminal network under various attack strategies, especially targeting its modular structure.
• To assess the network's controllability using exact network control theory to identify minimal control sets.
• To provide actionable insights for law enforcement agencies on proactive disruption strategies based on network topology.

Proposed method

• Construction of an undirected, unweighted network from law enforcement data on federal crimes in Brazil, with 23,666 individuals and 35,913 relationships.
• Application of the Louvain algorithm to detect community structure and compute modularity (Q=0.96), indicating strong community separation.
• Evaluation of attack strategies: random, degree-based, betweenness-based, and module-based attacks (MBA) to assess network fragility.
• Computation of driver nodes using the rank of the adjacency matrix: nD = 1/N * max{1, N - rank(A)} to determine minimal control set.
• Analysis of network properties including small-worldness, clustering coefficient, edge density, and network efficiency to characterize the network's 'dark' nature.
• Comparison of attack performance using the fraction of edges/nodes removed and the resulting fragmentation of the giant component.

Experimental results

Research questions

• RQ1 What is the topological structure of the Brazilian federal crime network, particularly in terms of community organization and connectivity?
• RQ2 How vulnerable is the network to targeted attacks, and which attack strategy—module-based or centrality-based—achieves disruption with minimal resource expenditure?
• RQ3 To what extent is the criminal network controllable in the mathematical sense, and how many driver nodes are required to control the entire system?
• RQ4 How do the network's structural properties (e.g., low density, high modularity) influence its resilience and vulnerability to disruption?

Key findings

• The network exhibits a giant component containing 40% of the nodes and 54% of the edges, indicating a highly interconnected core.
• The network has a very high modularity (Q=0.96), revealing a strong community structure that makes it highly vulnerable to module-based attacks.
• Only 2% of nodes or edges need to be removed via module-based attacks to fully fragment the giant component, demonstrating extreme topological fragility.
• The network is mathematically controllable with only 21% of nodes (2,076 out of 9,887 in the giant component) acting as driver nodes, consistent with typical social network behavior.
• Betweenness-based attacks are highly effective, but module-based attacks achieve full fragmentation with less computational cost and near-identical disruption efficiency.
• Despite mathematical controllability, practical implementation faces ethical, legal, and operational challenges due to the human nature of driver nodes and the dynamic nature of criminal systems.