Organised crime continues to be one of the most difficult issues for governments to combat worldwide, and Italy, home to some of the world’s most notorious criminal organisations like the Mafia, is no exception (Pinotti 2020). Organised crime thrives where governments are weak, stepping in to fill gaps in law enforcement and protection that the state fails to provide (Isopi et al. 2012). In these power vacuums, criminal groups grow stronger, embedding themselves in communities (De Feo et al. 2018). Beyond traditional illegal activities, they infiltrate the economy, redirect public funds, and undermine local development (Narciso and Barone 2012). The consequences are widespread: they distort markets, fuel corruption, and escalate violence (Sorrenti and Le Moglie 2020, Mocetti et al. 2019).
In Italy, the Mafia’s deep-rooted presence has had lasting effects, particularly in local governments, where criminal organisations infiltrate municipal bodies to control public resources and manipulate elections. Detecting this infiltration, however, has proven to be one of the most significant challenges for authorities. In a new study (Campedelli et al. 2024), we propose that machine learning could help change this.
Our study leverages machine learning to predict mafia infiltration in local governments, offering a tool that can identify at-risk municipalities before criminal involvement is formally detected by the state. This innovative approach holds the potential to improve the effectiveness of anti-mafia policies, prevent the misuse of public funds, and provide early warnings for local governments vulnerable to organised crime.
The challenge of detecting Mafia infiltration
The Italian government has long recognised the threat of mafia infiltration in local politics and has in turn designed policies to address it. One of the most assertive actions that can be taken is the dismissal of municipal governments found to be infiltrated by organised crime. This disruptive measure enables the central government to restore control over areas where the mafia has compromised democratic processes. However, this intervention comes only after mafia infiltration has already been identified, a process that is often slow, reactive, and dependent on visible evidence of criminal behaviour.
Detecting organised crime is particularly difficult due to the clandestine nature of mafia activities, such as corruption and money laundering, which leave little trace. Investigations typically rely on observable indicators, such as violent crime or corruption scandals, to expose mafia influence. This reactive approach allows organised crime to establish a firm foothold before the state can respond. In the meantime, mafias appropriate public resources, distort local governance, and strengthen their political and economic power.
Our study aims to address this issue by predicting infiltration before it becomes evident, shifting from a reactive to a proactive stance in combating organised crime.
The power of machine learning in predicting infiltration
To predict mafia infiltration, we collect data on Italian local elections and municipal budgets spanning 2001 to 2020. These data were used to develop a machine-learning framework that could forecast the likelihood of mafia involvement in local governments. The model developed using the XGBoost algorithm emerged as the most effective among the nine machine-learning techniques and over 200 architectures tested. Remarkably, the best model was able to predict up to 96% of out-of-sample local governments that the Italian state had already identified as infiltrated by organised crime – up to two years before detection. Figure 1 displays this index in three different years, showing substantial heterogeneities across and within regions.
Figure 1 Predicted mafia infiltration risk in Italy
Notes: The colormap has four colours: orange if the predicted risk is between 0 and the overall median value for the 2001–2020 period (i.e. 0.02); red if the risk lies between 0.02 and the value representing the 75th percentile for the entire period (i.e. 0.12); cyan if the risk lies between the 75th and the 95th percentile (i.e. 0.96); and blue if the risk is higher than the 95th percentile.
This predictive tool bears significant implications for law enforcement and policymakers. Not only can it help identify infiltrated municipalities sooner, but it can also flag ‘high-risk’ municipalities that may not have been detected with traditional methods. By focusing on the top 5% of municipalities deemed at risk in any given year, the model successfully predicted 89% of infiltrated governments. While this comes with a precision rate of about 15% – meaning one in six flagged municipalities is truly infiltrated – the high recall rate ensures that few infiltrated governments escape detection.
Importantly, the model’s false positives may still be valuable. Municipalities flagged by the model but not yet identified by the state may represent cases of infiltration that have simply evaded detection. We found that these municipalities often display similar characteristics to those already confirmed as infiltrated, such as experiencing mafia-related violence or attacks on local politicians. This suggests that false positives could indeed indicate municipalities at high risk of future infiltration.
The influence of public spending on Mafia infiltration
Another significant finding from the study is the link between public spending and mafia infiltration. Organised crime groups infiltrate local governments primarily to capture public resources, and we found that an influx of public money, such as EU transfers, can increase the risk of mafia involvement in local politics.
Here, we focused on the disbursement of EU structural funds to southern Italy between 2007 and 2013, a period that saw a reduction in fiscal oversight and a significant increase in public funding for local governments. Figure 2 shows the high correlation between EU funds disbursement and local governments infiltrated by organised crime.
Figure 2 EU funds disbursement and local government infiltration by organised crime (southern Italy)
Notes: This figure visualises the trends (from 2000 to 2017) in terms of yearly number of city council dismissals (red line) and EU funds assigned to Italy, in billion euros (blue line). The Pearson correlation between the two is 0.63 (p-val=0.005).
Using a geographic differences-in-discontinuities approach, we compared municipalities on either side of regional borders that received or did not receive these funds. The analysis shows that municipalities receiving EU funds saw an increase in the predicted risk of mafia infiltration by 11 to 14 percentage points. Remarkably, the effect emerges even when comparing municipalities that are located within 5 kilometres of the regional borders separating treated and control regions.
This finding raises important questions about the unintended consequences of public financial aid. While EU funds are designed to promote economic growth and reduce poverty in disadvantaged regions, they may also attract the attention of criminal organisations seeking to exploit weak oversight and capture public resources for their own gain. The study’s predictive model can serve as an early warning system, allowing policymakers to monitor high-risk municipalities more closely and implement stronger safeguards to prevent the misuse of public funds.
Broader implications and the future of crime prevention
The use of machine learning to predict mafia infiltration represents a significant step forward in the fight against organised crime. Rather than relying solely on traditional investigative methods, which often uncover criminal activity after it has taken root, law enforcement might use predictive models to stay ahead of organised crime. This proactive approach has the potential to reshape how governments respond to mafia infiltration and could lead to earlier interventions that prevent the misuse of public resources and preserve the integrity of local governments.
Moreover, this model could be adapted and applied in other regions facing similar challenges with organised crime. Countries in the Balkans, Eastern Europe, and Latin America – where corruption and organised crime are deeply intertwined with politics – could benefit from similar machine-learning tools.
References
Campedelli, G M, G Daniele and M Le Moglie (2024), “Mafia, politics and machine predictions”, CEPR Discussion Paper 19322.
De Feo, G, G De Luca, and D Acemoglu (2018), “Weak states: Causes and consequences of the Sicilian Mafia”, VoxEU.org, 2 March.
Isopi, A, O Olsson, and A Dimico (2012), “Origins of the Sicilian Mafia”, VoxEU.org, 26 July.
Narciso, G, and G Barone (2012), “Can the Mafia divert the allocation of public transfers?”, VoxEU.org, 5 May.
Mocetti, S, L Rizzica, and L Mirenda (2019) “The boss on board: Mafia infiltrations, firm performance, and local economic growth”, VoxEU.org, 26 October.
Sorrenti, G, and M Le Moglie (2020), “When godfathers become entrepreneurs: On the organised crime’s infiltration in legal economy”, VoxEU.org, 1 August.
Pinotti, P (2020), “Burden of proof: Measuring and understanding crime”, VoxEU.org, 1 August.
