DATA-DRIVEN INTELLIGENCE IN DRUG CRIME PREVENTION: PREDICTIVE ANALYTICS, CASE STUDIES, AND ETHICAL PERSPECTIVES

Abstract

Abstract

Drug-related crime presents complex public health, security, and governance challenges. This paper synthesizes evidence on predictive analytics for drug crime prevention, integrating insights from criminology, data science, and policy. We review the distinct characteristics of drug markets; describe data sources and analytic foundations (e.g., near-repeat models, risk terrain modeling, spatio-temporal forecasting); and examine applications ranging from hotspot policing to online trafficking intelligence and overdose early-warning systems. Case studies from the United States, Europe, and Asia illustrate real-world implementation, including successes and failures. We critically assess ethical and legal concerns—bias, privacy, transparency, accountability—and situate them within evolving regulatory frameworks, notably the EU AI Act’s prohibition of predictive policing and governance requirements for high-risk AI systems. We conclude with practical safeguards and research priorities to ensure that predictive tools complement comprehensive strategies emphasizing public health, prevention, and community trust.

Keywords: predictive policing; drug markets; overdose surveillance; fairness; EU AI Act; risk terrain modeling; near-repeat; spatio-temporal forecasting.

1. Introduction

Illicit drug markets undermine public safety and well-being: trafficking fuels violence, corrodes institutions via corruption, and strains health systems (EMCDDA & Europol, 2024). Their scale and adaptability – new substances, shifting routes, rapid online coordination – challenge conventional enforcement. In response, many jurisdictions have experimented with data-driven methods, broadly referred to as predictive policing, to anticipate crime patterns and deploy resources proactively (Perry et al., 2013; Storbeck, 2022).

We first outline what makes drug crime different and why that matters for analytics. We then review analytic foundations and data sources, followed by application areas and illustrative case studies. We evaluate evidence of impact and examine ethical–legal implications, including fairness trade-offs (Kleinberg et al., 2017) and systemic bias risks (Barocas & Selbst, 2016; Lum & Isaac, 2016). We also connect policing to public health, highlighting overdose surveillance (ODMAP) and treatment-oriented interventions. Finally, we discuss governance and future directions under emerging regulatory frameworks like the EU AI Act (Regulation (EU) 2024/1689).

Predictive policing was a “magic word” for a long time. Many people expected a miracle from it and considered it a crystal ball that could tell the future with 100% accuracy. Unfortunately, this is not the case, which is why many people turned away from it (Mátyás et al. 2025).

The first predictive policing software was created in Hungary in 2004 and was able to predict five types of crimes (car breaking, car theft, robbery, burglary, tricky theft), followed later by American software, and then the Italian KeyCrime software. There was a great breakthrough in the field in the 2010s. Hundreds of software were created that could predict expected crimes with varying degrees of efficiency. Software was used to predict more and more different types of crimes, so many software were able to predict drug-related crimes as well. It seemed that predictive policing could revolutionize policing. Unfortunately, this did not happen.

In the late 2010s and early 2020s, human rights activists fought against the use of the software, and its use was discontinued in many countries. Even the first and largest American software, PredPol (later Analitica), suffered this fate. It was discontinued in Los Angeles a few years ago. According to human rights activists, the software performs ethnic profiling and operates in a discriminatory manner.

The authors have been working on the criminology of drug crime for years, as well as predictive policing, and in 2024 they even participated in the development of an accident prediction software. This inspired them to address the question – for now only in principle – of how drug crime can be predicted.

This study is the first stage of a research, laying the theoretical foundations. The second level of the research will be the review and processing of several hundred drug-related criminal cases in Hungary. Based on these, they would like to create an algorithm that may be suitable for predicting drug-related crimes.

Many modern predictive software uses camera recordings for predictions, which refine the probability of the prediction (Vári et al. 2025). If the authors receive permission to use camera recordings, they also intend to use them in the analysis. Many geographical factors also play a major role in the commission of drug-related crimes – e.g. settlement structure (Bói 2024) –, hot spot analysis (Vajda 2024), etc. All of these factors must be taken into account when making the prediction.

As mentioned above, the role of predictive analytics has decreased in many Western countries. On the other hand, there are several countries (e.g. China, India) where the role of predictive software is not only decreasing, but also increasing year by year.

According to the authors, a software cannot be considered racist, but obviously human rights, the right to human dignity, etc. must be taken seriously, and of course no one should be discriminated against based on skin color, origin, language, religion, etc. With the appropriate use of algorithms, various predictive software has a place in law enforcement.