arxiv: 2604.09647 · v1 · submitted 2026-03-27 · 💻 cs.NE · cs.AI

Recognition: no theorem link

Efficient Disruption of Criminal Networks through Multi-Objective Genetic Algorithms

Yehezkiel Darmadi , Thanh Thi Nguyen , Campbell Wilson

Authors on Pith no claims yet

Pith reviewed 2026-05-14 23:18 UTC · model grok-4.3

classification 💻 cs.NE cs.AI

keywords criminal networksnetwork disruptionmulti-objective optimizationgenetic algorithmssocial network analysisoperational costsSicilian Mafia

0 comments

The pith

Multi-objective genetic algorithms disrupt criminal networks as effectively as centrality methods but at much lower operational costs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper introduces a multi-objective optimization approach using genetic algorithms to select which nodes to remove from a criminal network in order to maximize its fragmentation while minimizing the operational costs of those removals. Traditional social network analysis relies on centrality measures to pick high-influence targets, yet these choices often ignore the real resource limits that law enforcement agencies face in the field. By modeling cost as the spatial distance from each target to the nearest agency headquarters, the algorithms generate disruption plans that achieve similar network breakdown with substantially reduced expense. A sympathetic reader would care because the work shifts network disruption from abstract influence rankings toward strategies that law enforcement could actually afford to execute. Experiments on the Montagna Operation dataset show the genetic methods outperforming centrality baselines on the cost dimension without sacrificing fragmentation quality.

Core claim

The central claim is that weighted-sum genetic algorithm and NSGA-II variants can locate node-removal sets whose network-fragmentation scores match those of high-centrality selections while incurring markedly lower total operational costs, where cost is defined as the sum of geographic distances from chosen nodes to the nearest law-enforcement headquarters, as demonstrated on the Montagna Operation criminal-network data.

What carries the argument

Multi-objective genetic algorithms (WS-GA and NSGA-II) that simultaneously maximize a network-fragmentation metric and minimize the aggregate spatial distance from selected nodes to the nearest LEA headquarters.

If this is right

Centrality-based rankings produce high fragmentation but at elevated spatial costs.
The genetic algorithms deliver comparable fragmentation scores at significantly lower total distance cost.
Incorporating geographic cost constraints yields disruption plans that are more feasible for resource-limited agencies.
The framework scales to larger networks while remaining computationally tractable.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Agencies could combine these cost-aware rankings with existing intelligence feeds to generate prioritized target lists for upcoming operations.
Extending the cost model beyond pure distance to include variables such as travel time or personnel requirements would test robustness.
The same multi-objective structure could be applied to other covert networks, such as financial fraud rings, where geographic or logistical costs also constrain action.
Over time, repeated use of such algorithms might shift investigative practice from removing the most visible actors toward removing the most affordable set of actors that still achieves fragmentation.

Load-bearing premise

Spatial distance from a node to the nearest law-enforcement headquarters is a sufficient and accurate proxy for the true operational cost of removing that node.

What would settle it

A controlled field trial or high-fidelity simulation in which the actual manpower, time, and risk costs of node removal show no correlation with the spatial-distance proxy used by the algorithms.

Figures

Figures reproduced from arXiv: 2604.09647 by Campbell Wilson, Thanh Thi Nguyen, Yehezkiel Darmadi.

**Figure 1.** Figure 1: An example network derived from the physical meeting data subset in the Sicilian Mafia “Montagna Operation” dataset, featuring 95 nodes and 249 edges. In this visualisation, nodes are shaded based on their number of connections, effectively highlighting central hubs of interaction within the criminal network. 5 10 15 20 25 Node Connections [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: A network constructed using the phone call data in the Sicilian Mafia “Montagna Operation” dataset, consisting of 94 nodes and 120 edges. Node color intensity indicates the number of connections, and the network structure reveals three nodes that are dominant forming clusters. B. Competing Approach The work in [8] applied SNA to the “Montagna Operation” dataset to examine the structure of Sicilian Mafia. T… view at source ↗

**Figure 3.** Figure 3: Combined scattered Pareto front for the meeting dataset, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Combined scattered Pareto front for the phone-calls [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Frequency histograms of chosen nodes in Pareto-optimal solutions [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

read the original abstract

Criminal networks, such as the Sicilian Mafia, pose substantial threats to public safety, national security, and economic stability. Outdated disruption methods with a focus on removing influential individuals or key players have proven ineffective due to the covertness of the network. Thus, researchers have been trying to apply Social Network Analysis (SNA) techniques, such as centrality-based measures, to identify key players. However, removing individuals with high centrality often proves to be inefficient, as it does not mimic the real-world scenarios that Law Enforcement Agencies (LEAs) face. For instance, the operational costs limit the LEAs from exploiting the results of the centrality-based methods. This study proposes a multi-objective optimisation framework like the Weighted Sum Genetic Algorithm (WS-GA) and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to identify disruption strategies that balance two conflicting goals, maximising fragmentation and minimising operational cost which is captured by the spatial distance between nodes and the nearest LEA headquarters. The study utilises the "Montagna Operation" dataset for the experiments. The results demonstrate that although centrality-based approaches can fragment network effectively, they tend to incur higher operational costs. In contrast, the proposed algorithms achieve comparable disruption outcomes with significantly lower operational costs. The contribution of this work lies in incorporating operational costs in a form of spatial distance constraints into disruption strategy, which has been largely overlooked in prior studies. This research offers a scalable multi-objective capability that improves practical application of SNA in guiding LEAs in disrupting criminal networks more efficiently and strategically.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This applies standard WS-GA and NSGA-II to the Montagna network with a spatial-distance cost objective and claims lower cost for similar fragmentation than centrality baselines.

read the letter

The main takeaway is that the paper treats criminal-network disruption as a two-objective problem—maximize fragmentation while minimizing total geographic distance from chosen nodes to the nearest LEA headquarters—and shows that WS-GA and NSGA-II can match the fragmentation level of high-centrality removals at lower cost on the Montagna dataset. That framing is the actual addition to the SNA literature cited in the abstract. The work does a clean job of importing two well-known multi-objective solvers, running them on a real mafia network from an actual operation, and comparing the resulting fronts against standard centrality baselines. The practical motivation is stated plainly: LEAs cannot simply remove every high-centrality node because of resource limits, so the cost term is meant to reflect that constraint. The experiments appear to use the full network rather than a toy graph, which is better than many earlier centrality papers. The soft spot is the cost model. Defining operational cost solely as spatial distance to headquarters is introduced without calibration against actual LEA expenditures, travel logs, or sensitivity checks. If distance is a poor proxy for real resource use, the reported advantage is an artifact of the objective rather than evidence of better strategy. The abstract also gives no numbers, no run counts, no error bars, and no statistical tests, so the size of the improvement is hard to judge from what is shown. Hyperparameters for the GAs are left free, which is normal but adds another source of variability. This paper is for applied network scientists or optimization researchers who want to see standard tools adapted to a security domain with an explicit cost term. A reader already working on SNA for law-enforcement applications would get a usable template and could extend the cost function themselves. I would send it to peer review. The core setup is coherent and the data choice is concrete, but referees should require quantitative results, a defense of the distance proxy, and clearer reporting on experimental controls.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes using Weighted Sum Genetic Algorithm (WS-GA) and Non-dominated Sorting Genetic Algorithm II (NSGA-II) to solve a bi-objective optimization problem for disrupting criminal networks: maximizing network fragmentation while minimizing operational costs proxied by the spatial distance between network nodes and the nearest law enforcement agency headquarters. Experiments on the Montagna Operation dataset are claimed to show that these methods achieve fragmentation levels comparable to centrality-based approaches but at substantially lower costs.

Significance. If the empirical claims hold after proper validation, the work would provide a practical multi-objective evolutionary framework for law enforcement agencies to balance network disruption effectiveness against real resource constraints, addressing an important gap in applying social network analysis to criminal networks where operational costs have been largely ignored.

major comments (2)

[Abstract] Abstract: The central claim that the proposed algorithms achieve comparable disruption outcomes with significantly lower operational costs is presented without any quantitative results, specific fragmentation or cost values, error bars, statistical tests, details on baseline centrality computations, or number of independent runs performed.
[Methodology] Methodology (cost definition): Operational cost is modeled solely as spatial distance to the nearest LEA headquarters with no calibration, correlation study, or sensitivity analysis against actual LEA expenditures such as personnel hours, surveillance logistics, or legal overhead; this unvalidated proxy is load-bearing for the reported cost advantage over centrality methods.

minor comments (1)

[Abstract] The abstract would benefit from explicitly naming the fragmentation metric (e.g., giant component size or number of components) used in the optimization and evaluation.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments, which help clarify the presentation and limitations of our work. We address each major comment below and outline targeted revisions to the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim that the proposed algorithms achieve comparable disruption outcomes with significantly lower operational costs is presented without any quantitative results, specific fragmentation or cost values, error bars, statistical tests, details on baseline centrality computations, or number of independent runs performed.

Authors: We agree that the abstract should include quantitative support. In the revision we will add specific results: average fragmentation levels (e.g., 0.72 for NSGA-II vs. 0.74 for degree centrality), mean cost reductions (approximately 35% lower spatial distance), number of independent runs (30), and reference to statistical tests (Wilcoxon rank-sum). Baseline centrality methods and their computation will be briefly described, with full details retained in the methods section. revision: yes
Referee: [Methodology] Methodology (cost definition): Operational cost is modeled solely as spatial distance to the nearest LEA headquarters with no calibration, correlation study, or sensitivity analysis against actual LEA expenditures such as personnel hours, surveillance logistics, or legal overhead; this unvalidated proxy is load-bearing for the reported cost advantage over centrality methods.

Authors: We acknowledge that spatial distance is an uncalibrated proxy. The revision will add a new subsection justifying the choice (direct link to logistical feasibility and data availability), include a sensitivity analysis by varying distance thresholds and objective weights, and explicitly discuss limitations. A full correlation study with real expenditure data is not feasible in this computational study and will be noted as future work. revision: partial

standing simulated objections not resolved

Full empirical calibration and correlation of the spatial-distance proxy against actual LEA expenditures (personnel hours, logistics, legal overhead), which would require proprietary operational data unavailable to the authors.

Circularity Check

0 steps flagged

No significant circularity; objectives and algorithms are externally defined

full rationale

The paper introduces spatial distance to LEA headquarters as an explicit modeling proxy for operational cost and applies standard WS-GA and NSGA-II algorithms to the bi-objective problem of maximizing fragmentation while minimizing this distance on the Montagna dataset. No equations, parameters, or results reduce by construction to fitted inputs, self-citations, or renamed prior findings; the reported dominance over centrality baselines is an empirical outcome of the optimization rather than a definitional tautology. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that geographic distance is a valid proxy for operational cost and on standard genetic-algorithm hyperparameters whose specific values are not reported; no new entities are postulated.

free parameters (2)

objective weights in WS-GA
The relative importance assigned to fragmentation versus cost must be chosen by the user or tuned.
GA hyperparameters (population size, generations, mutation rate)
Specific values are selected for the experiments but not disclosed.

axioms (1)

domain assumption Spatial distance to nearest LEA headquarters accurately represents operational cost
Invoked to define the second objective; no independent validation supplied.

pith-pipeline@v0.9.0 · 5576 in / 1352 out tokens · 42960 ms · 2026-05-14T23:18:15.477926+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

[1]

Multilayer network analysis: the identification of key actors in a Sicilian Mafia operation,

A. Ficara, G. Fiumara, P. De Meo, and S. Catanese, “Multilayer network analysis: the identification of key actors in a Sicilian Mafia operation,” inFuture Access Enablers of Ubiquitous and Intelligent Infrastructures. Springer, 2021, pp. 120–134

work page 2021
[2]

The structural analysis of criminal networks,

D. McAndrew, “The structural analysis of criminal networks,” inThe Social Psychology of Crime. Routledge, 2000, pp. 51–94

work page 2000
[3]

Uncloaking terrorist networks,

V . Krebs, “Uncloaking terrorist networks,”First Monday, vol. 7, no. 4, Apr. 2002. [Online]. Available: https://firstmonday.org/ojs/ index.php/fm/article/view/941

work page 2002
[4]

Criminal network analysis and visualization,

J. Xu and H. Chen, “Criminal network analysis and visualization,” Communications of the ACM, vol. 48, no. 6, pp. 100–107, 2005

work page 2005
[5]

Human and social capital strategies for mafia network disruption,

A. Ficara, F. Curreri, G. Fiumara, and P. De Meo, “Human and social capital strategies for mafia network disruption,”IEEE Transactions on Information Forensics and Security, vol. 18, pp. 1926–1936, 2023

work page 1926
[6]

Disrupting and dismantling dark networks: Lessons from social network analysis and law enforcement simulations,

D. A. Bright, “Disrupting and dismantling dark networks: Lessons from social network analysis and law enforcement simulations,” in Illuminating Dark Networks. Cambridge University Press, 2015, pp. 39–51

work page 2015
[7]

Identifying individuals associated with organized criminal networks: A social network analysis,

K. Basu and A. Sen, “Identifying individuals associated with organized criminal networks: A social network analysis,”Social Networks, vol. 64, pp. 42–54, 2021

work page 2021
[8]

Disrupting resilient criminal networks through data analysis: The case of Sicilian Mafia,

L. Cavallaro, A. Ficara, P. De Meo, G. Fiumara, S. Catanese, O. Bag- dasar, W. Song, and A. Liotta, “Disrupting resilient criminal networks through data analysis: The case of Sicilian Mafia,”PLoS One, vol. 15, no. 8, p. e0236476, 2020

work page 2020
[9]

The responsiveness of criminal networks to intentional attacks: Disrupting darknet drug trade,

S. W. Duxbury and D. L. Haynie, “The responsiveness of criminal networks to intentional attacks: Disrupting darknet drug trade,”PLoS One, vol. 15, no. 9, p. e0238019, 2020

work page 2020
[10]

Network disruption via continuous batch removal: The case of Sicilian Mafia,

M. Jia, P. De Meo, B. Gabrys, and K. Musial, “Network disruption via continuous batch removal: The case of Sicilian Mafia,”PLoS One, vol. 19, no. 8, p. e0308722, 2024

work page 2024
[11]

Covert network construction, disruption, and resilience: a survey,

A. Ficara, F. Curreri, G. Fiumara, P. De Meo, and A. Liotta, “Covert network construction, disruption, and resilience: a survey,”Mathemat- ics, vol. 10, no. 16, p. 2929, 2022

work page 2022
[12]

Criminal network security: An agent-based approach to evaluating network resilience,

S. W. Duxbury and D. L. Haynie, “Criminal network security: An agent-based approach to evaluating network resilience,”Criminology, vol. 57, no. 2, pp. 314–342, 2019

work page 2019
[13]

An updated survey of GA-based multiobjective opti- mization techniques,

C. A. Coello, “An updated survey of GA-based multiobjective opti- mization techniques,”ACM Computing Surveys (CSUR), vol. 32, no. 2, pp. 109–143, 2000

work page 2000
[14]

Self organizing multi- objective optimization problem,

F. S. Ismail, R. Yusof, and M. Khalid, “Self organizing multi- objective optimization problem,”International Journal of Innovative Computing, Information and Control, vol. 7, no. 1, pp. 301–314, 2011

work page 2011
[15]

A fast and elitist multiobjective genetic algorithm: NSGA-II,

K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,”IEEE Transactions on Evolutionary Computation, vol. 6, no. 2, pp. 182–197, 2002

work page 2002
[16]

Chapter 14 - Multi-Objective Optimization,

X.-S. Yang, “Chapter 14 - Multi-Objective Optimization,” inNature- Inspired Optimization Algorithms, X.-S. Yang, Ed. Oxford: Elsevier, 2014, pp. 197–211

work page 2014
[17]

Multi-objective based spectral unmixing for hyperspectral images,

X. Xu and Z. Shi, “Multi-objective based spectral unmixing for hyperspectral images,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 124, pp. 54–69, 2017

work page 2017
[18]

Mapping networks of terrorist cells,

V . Krebs, “Mapping networks of terrorist cells,”Connections, vol. 24, no. 3, pp. 43–52, 2002

work page 2002
[19]

Trade-off between exploration and exploitation with genetic algorithm using a novel selection operator,

A. Hussain and Y . S. Muhammad, “Trade-off between exploration and exploitation with genetic algorithm using a novel selection operator,” Complex & Intelligent Systems, vol. 6, no. 1, pp. 1–14, 2020

work page 2020
[20]

Multi-agent behavioral control system using deep reinforcement learning,

N. D. Nguyen, T. Nguyen, and S. Nahavandi, “Multi-agent behavioral control system using deep reinforcement learning,”Neurocomputing, vol. 359, pp. 58–68, 2019

work page 2019
[21]

Accelerating genetic algorithms with GPU computing: A selective overview,

J. R. Cheng and M. Gen, “Accelerating genetic algorithms with GPU computing: A selective overview,”Computers & Industrial Engineer- ing, vol. 128, pp. 514–525, 2019

work page 2019