AI Predicts Crime Hotspots: Building Safer Cities with Smarter Data
In a world of limited resources—whether budget, personnel, or time—crimes continue to cluster in certain neighborhoods. This raises an important question: Can we predict where crimes are likely to happen before they occur?
A recent study by Zubair et al. (2025) suggests that we can. The researchers developed a cutting-edge AI model that predicts crime hotspots with impressive accuracy using a technique called Graph Convolutional Networks (GCNs).
By analyzing over 7 million crime records from Chicago, collected since 2001, their AI learned patterns of criminal activity, ultimately achieving 88% accuracy in predicting high-risk areas.
Traditional crime forecasting methods—like Kernel Density Estimation (KDE) or Support Vector Machines (SVM)—have typically analyzed crime locations in isolation, failing to account for how crime in one area can affect nearby neighborhoods. GCNs overcome this limitation by treating the city as an interconnected network, recognizing that crime often spreads from one area to its surroundings.
In practice, the AI divides the city into 2.2 x 2.2 km grids, linking adjacent areas and analyzing how past crime patterns evolve over space and time. The result? Not just high accuracy, but also an impressive F1-score of 0.83, meaning the predictions strike a good
balance between precision and coverage.
Figure 1: An excerpt of graph
The AI can even generate heatmaps that visually highlight where risks are concentrated, making it easier for practitioners to understand and act.
In Chicago, these insights have moved beyond research papers. They’re being applied to help police plan patrol routes, decide where to install CCTV cameras, improve lighting, and even redesign public spaces to deter crime. This data-driven approach also helps build public trust by showing that safety measures are based on solid evidence—not just intuition or guesswork.
Figure 2: Heat map for crime type “Theft (Left)” and "Narcotics (Right)"
Could this work in Thailand?
Absolutely—and it should. Thailand faces similar urban crime challenges, from theft and assault to violence in vulnerable communities. The key challenge is that Thailand currently lacks the kind of detailed spatial crime data that makes these AI tools work effectively.
A good starting point would be to implement this AI in cities where data and infrastructure are already in place—such as Bangkok, Pattaya, and Chiang Mai—all complex urban environments with significant resident and tourist populations. With systematic data collection, these cities could serve as ideal pilot sites.
But technology and data alone won’t be enough. What’s essential is real-world implementation and collaboration between policy-makers, law enforcement, researchers, and communities.
👮 ♂️For police agencies, especially Metropolitan Police Bureau and metropolitan police stations—AI predictions could help deploy officers more efficiently, ensuring a strong and visible police presence in the areas that need it most.
🏛️ For city governments, data-driven insights would guide investments in lighting, public walkways, CCTV, and neighborhood improvements—tailoring interventions to the specific needs of each community.
🎓 For universities and researchers, the challenge and opportunity lie in developing locally-relevant crime science models—integrating technical expertise with an understanding of Thailand’s unique social contexts, rather than simply copying foreign solutions.
👥 For citizens and communities, this information can empower people to take part in designing safer neighborhoods and foster a shared sense of responsibility for public safety.
When every sector comes together around data-driven decision-making, the goal of a "safe city" moves from being just a vision or a plan on paper to becoming a real, tangible improvement in people’s daily lives.
Reference:
Zubair, T., Fatima, S. K., Ahmed, N., & Khan, A. (2025). Crime Hotspot Prediction Using Deep Graph Convolutional Networks. arXiv preprint arXiv:2506.13116.
#KnowledgeHub #CenterForCrimeScience #CrimeScience #CrimePrediction #AI #SmartCity #SafeCity
