Predicting crime at micro places : comparing machine learning methods across European cities
- Author
- Robin Khalfa (UGent) , Thom Snaphaan (UGent) , Alina Ristea, Ourania Kounadi and Wim Hardyns (UGent)
- Organization
- Project
- Abstract
- The present study compares the performance of three different supervised machine learning methods, namely an Ensemble Neural Network algorithm (ENN), a Random Forest algorithm (RF), and a K-Nearest Neighbor algorithm (KNN), in predicting residential burglary hot spots across different cities in Europe, i.e., Brussels, Vienna and London. Crime and crime-supporting data are collected for the three cities, spanning the period 2014-2016. The data are (dis)aggregated to a 200x200m grid overlaying the study areas and monthly predictions are made for each month in 2016 using the so-called rolling window approach. For each method and city, four prediction performance measures are calculated and compared (i.e., direct hit rate, near hit rate, precision and F1-scores). The results indicate that the ENN and RF algorithm achieve comparable prediction performance when predicting a smaller number of high-risk grid cells, outperforming the KNN algorithm. This suggests that in general, law enforcement agencies wishing to apply a spatiotemporal crime prediction approach should be cognizant of the enhanced performance exhibited by ensemble machine learning models such as the ENN and RF compared to non-ensemble methods such as KNN. Moreover, although the three algorithms achieved more consistent performance measures for London, no substantial differences in performance were observed across cities, suggesting that the predictive modeling approach used in this study holds premise for cross-city and cross-country application. The implications and limitations of this study are furthermore discussed.
- Keywords
- Crime prediction, Predictive modeling, Machine learning, Spatiotemporal crime analysis, Big data policing, Algorithm
Downloads
-
Khalfa et al. - Predicting crime at micro places - Accepted manuscript.pdf
- full text (Accepted manuscript)
- |
- open access
- |
- |
- 735.17 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01J5P45BQMH3B8RW1T3A6A162Q
- MLA
- Khalfa, Robin, et al. “Predicting Crime at Micro Places : Comparing Machine Learning Methods across European Cities.” New Research in Crime Modeling and Mapping Using Geospatial Technologies, edited by Michael Leitner, vol. 27, Springer, 2025, pp. 81–111, doi:10.1007/978-3-031-81580-5_5.
- APA
- Khalfa, R., Snaphaan, T., Ristea, A., Kounadi, O., & Hardyns, W. (2025). Predicting crime at micro places : comparing machine learning methods across European cities. In M. Leitner (Ed.), New research in crime modeling and mapping using geospatial technologies (Vol. 27, pp. 81–111). https://doi.org/10.1007/978-3-031-81580-5_5
- Chicago author-date
- Khalfa, Robin, Thom Snaphaan, Alina Ristea, Ourania Kounadi, and Wim Hardyns. 2025. “Predicting Crime at Micro Places : Comparing Machine Learning Methods across European Cities.” In New Research in Crime Modeling and Mapping Using Geospatial Technologies, edited by Michael Leitner, 27:81–111. Springer. https://doi.org/10.1007/978-3-031-81580-5_5.
- Chicago author-date (all authors)
- Khalfa, Robin, Thom Snaphaan, Alina Ristea, Ourania Kounadi, and Wim Hardyns. 2025. “Predicting Crime at Micro Places : Comparing Machine Learning Methods across European Cities.” In New Research in Crime Modeling and Mapping Using Geospatial Technologies, ed by. Michael Leitner, 27:81–111. Springer. doi:10.1007/978-3-031-81580-5_5.
- Vancouver
- 1.Khalfa R, Snaphaan T, Ristea A, Kounadi O, Hardyns W. Predicting crime at micro places : comparing machine learning methods across European cities. In: Leitner M, editor. New research in crime modeling and mapping using geospatial technologies. Springer; 2025. p. 81–111.
- IEEE
- [1]R. Khalfa, T. Snaphaan, A. Ristea, O. Kounadi, and W. Hardyns, “Predicting crime at micro places : comparing machine learning methods across European cities,” in New research in crime modeling and mapping using geospatial technologies, vol. 27, M. Leitner, Ed. Springer, 2025, pp. 81–111.
@incollection{01J5P45BQMH3B8RW1T3A6A162Q,
abstract = {{The present study compares the performance of three different supervised machine learning methods, namely an Ensemble Neural Network algorithm (ENN), a Random Forest algorithm (RF), and a K-Nearest Neighbor algorithm (KNN), in predicting residential burglary hot spots across different cities in Europe, i.e., Brussels, Vienna and London. Crime and crime-supporting data are collected for the three cities, spanning the period 2014-2016.
The data are (dis)aggregated to a 200x200m grid overlaying the study areas and monthly predictions are made for each month in 2016 using the so-called rolling window approach. For each method and city, four prediction performance measures are calculated and compared (i.e., direct hit rate, near hit rate, precision and F1-scores). The results indicate that the ENN and RF algorithm achieve comparable prediction performance when predicting a smaller number of high-risk grid cells, outperforming the KNN algorithm. This suggests that in general, law enforcement agencies wishing to apply a spatiotemporal crime prediction approach should be cognizant of the enhanced performance exhibited by ensemble machine learning models such as the ENN and RF compared to non-ensemble methods such as KNN. Moreover, although the three algorithms achieved more consistent performance measures for London, no substantial differences in performance were observed across cities, suggesting that the predictive modeling approach used in this study holds premise for cross-city and cross-country application. The implications and limitations of this study are furthermore discussed.}},
author = {{Khalfa, Robin and Snaphaan, Thom and Ristea, Alina and Kounadi, Ourania and Hardyns, Wim}},
booktitle = {{New research in crime modeling and mapping using geospatial technologies}},
editor = {{Leitner, Michael}},
isbn = {{9783031815799}},
issn = {{2365-0575}},
keywords = {{Crime prediction,Predictive modeling,Machine learning,Spatiotemporal crime analysis,Big data policing,Algorithm}},
language = {{eng}},
pages = {{81--111}},
publisher = {{Springer}},
series = {{Geotechnologies and the Environment}},
title = {{Predicting crime at micro places : comparing machine learning methods across European cities}},
url = {{http://doi.org/10.1007/978-3-031-81580-5_5}},
volume = {{27}},
year = {{2025}},
}
- Altmetric
- View in Altmetric