From the outset, BIGDATPOL has positioned itself as more than a research project: it is also a hub for dialogue between science, policy, and practice. The programme actively builds international collaborations with universities and research centres, and is strongly embedded in professional and policy networks at both the European and global level. These connections ensure that BIGDATPOL’s work resonates beyond academia and feeds directly into ongoing policy debates.
Public engagement has been equally central to the project’s identity. BIGDATPOL has been highlighted as a flagship initiative within Ghent University, invited to present results in leading European forums, and recognised in public-facing events and media. Such visibility has stimulated debate on the ethical and societal implications of data-driven policing and reinforced the programme’s role as a reference point for transparency and legitimacy.
BIGDATPOL PUBLICATIONS
Professor Wim Hardyns and his team have been studying big data policing for several years. A list of relevant publications can be found on Ghent University’s Biblio repository.
BIGDATPOL PRESENTATIONS
The BIGDATPOL project has been presented at various conferences and study days. A list of related outputs is available on Biblio.
HIGHLIGHTED PUBLICATION
‘Comparing XAI techniques for interpreting short-term burglary predictions at micro-places’
We published an academic article in May 2025 on a comparative study of XAI techniques for interpreting short-term burglary predictions. This article appeared in Computational Urban Science and was authored by Robin Khalfa, Naomi Theinert and Wim Hardyns.
This study empirically compares multiple eXplainable Artificial Intelligence (XAI) techniques to interpret short-term (weekly) machine learning-based burglary predictions at the micro-place level in Ghent, Belgium. While previous research predominantly relies on SHAP to interpret spatiotemporal crime predictions, this is the first study to systematically evaluate SHAP alongside other XAI techniques, offering both global and local model interpretability within the context of crime prediction.
What is SHAP?
SHAP (SHapley Additive exPlanations) is an Explainable AI (XAI) technique that helps interpret complex machine learning models by showing how much each feature contributes to a specific prediction.
It is based on Shapley values from game theory, which fairly distribute a model’s prediction among all input features, making it possible to understand both global patterns and individual (local) predictions.
Using data from 2014 to 2018 on residential burglary, repeat and near-repeat victimisation, environmental features, socio-demographic indicators, and seasonal effects, the researchers trained an XGBoost model with 76 features to predict weekly burglary hot spots. This model served as a basis for comparing the interpretative power of different XAI techniques. Results show that built environment and land use characteristics are the most consistent global predictors of burglary risk.
However, their influence varies substantially at the local level, revealing the importance of spatial context. While global feature importance rankings are broadly aligned across XAI techniques, local explanations, especially between SHAP and LIME, often diverge. These discrepancies highlight the need for careful method selection when translating predictions into crime prevention strategies.
In addition, this study demonstrates that short-term burglary risks are influenced by complex interactions and threshold effects between environmental and social disorganisation features. The findings are interpreted through criminological theory, calling for integrated approaches that go beyond examining isolated crime predictors.
The study concludes with reflections on the methodological implications of applying different interpretability techniques in predictive policing contexts.
Here are four key conclusions from the study:
- Simpler XAI techniques can replace SHAP for global insights, but local interpretations diverge. Across methods, the top global predictors of burglary risk were largely consistent, suggesting that computationally lighter techniques—such as permutation-based importance or ALE—may serve as pragmatic alternatives to SHAP when time and resources are limited. However, at the local level, SHAP and LIME sometimes yielded different explanations for the same prediction. These discrepancies could lead to diverging policy recommendations for targeted interventions, underlining that the choice of interpretability method matters most when models are used to guide operational decision-making.
- Burglary risk is shaped by complex, non-linear interactions between environmental and social factors. Burglary risk does not increase in a simple, linear way with any single factor. Instead, features such as street connectivity, population density, and socioeconomic disadvantage interact in ways that create threshold effects. For example, high street connectivity may only increase risk when coupled with low residential presence or economic hardship. These findings emphasise that prevention strategies should be tailored to local conditions—for instance, by enhancing natural surveillance or strengthening guardianship in highly connected but socially vulnerable areas.
- Predictive insights may vary across places, contexts, and crime types—and raise ethical questions. Because the study was conducted in Ghent, its findings may not generalise to cities with different street layouts, population structures, or social policies. For example, the influence of ethnic diversity on burglary risk may differ depending on local histories of migration and integration. The inclusion of sensitive socio-demographic variables in predictive models also raises ethical concerns about potential bias and fairness. Future research should test the robustness and equity of predictive interpretations across different contexts, timeframes, and crime types.
- Towards an integrated criminological perspective linking environment and social processes. The study underscores the need to move beyond analysing isolated risk factors and toward a more integrated understanding of how environmental design and social dynamics jointly shape opportunities for crime. Combining insights from crime opportunity theory and social disorganisation theory can help develop more context-sensitive, location-specific crime prevention strategies—grounded in both data and theory.