Predicting crime at micro places: Comparing different prediction methods across European cities

• In Event: Advances in quantitative methods 3

Abstract

The study evaluates and compares the performance of different machine learning models in predicting crime hotspots across three major European cities - Brussels, London, and Vienna. The predictive models utilize comprehensive datasets containing residential burglary data and crime-related features such as socio-demographic, socio-economic, environmental, proximity, and seasonal features spanning the period 2014-2016. The data are aggregated at a fine-grained spatial resolution of grid cells of 200 by 200-meters, and predictions are made for each month in 2016 using a rolling window approach. Accordingly, the dataset for each respective city is split into a training dataset, which is used to train the model including 10-fold cross-validation on a validation dataset, and a test dataset, which is used to evaluate how well the model is able to predict high-risk grid cells (i.e., hot spots). Four machine learning algorithms are employed in this study, namely an Ensemble Neural Network (ENN), a Random Forest (RF), a Support Vector Machine (SVM), and K-Nearest Neighbour (KNN). The modelling performance derived from these algorithms is evaluated using different established performance measures, including the direct hit rate, the near-hit rate, the precision, and the F1 score. This work contributes to the spatiotemporal study of crime by providing empirical evidence on the comparative performance of machine learning models in predicting crime hotspots across major European cities. It also sheds light on the role of local context in the efficacy of crime prediction models. The implications of the findings are multifaceted and may impact future research, policy formulation, and practical applications of spatiotemporal crime prediction in different urban contexts.

Authors

• Robin Khalfa, PhD Researcher, Institute for International Research on Criminal Policy, Department of Criminology, Criminal Law and Social Law, Ghent University

• Thom Snaphaan, PhD Researcher, Institute for International Research on Criminal Policy, Department of Criminology, Criminal Law and Social Law, Ghent University

• Ourania Kounadi, Assistant Professor Department for Geography and Regional Research, University of Vienna

• Alina Ristea, Assistant Professor Department of Security and Crime Science, University College London

• Wim Hardyns, Professor, Institute for International Research on Criminal Policy, Department of Criminology, Criminal Law and Social Law, Ghent University