Predicting crime in space and time: Examining and understanding input, throughput and output

Session Submission Type: Pre-arranged Panel

Abstract

Spatiotemporal crime prediction models offer valuable insights for crime prevention and public safety by enabling data-driven decision-making and resource allocation. However, key challenges remain, particularly regarding the reliability of crime data, the trade-off between model performance and interpretability, and the practical applicability of crime predictions/forecast within the context of operational and strategic decision-making processes. This panel brings together research that critically examines these challenges from multiple perspectives.

First, it explores the dynamic nature of (reported) crime data, analyzing how reporting processes influence data stability and (may) affect real-time crime predictions (input). Second, it examines how the features used in crime prediction models shape model decision-making, with a particular focus on how explainable artificial intelligence (XAI) techniques can enhance model interpretability (throughput). Third, it investigates the trade-offs between prediction performance and interpretability, particularly within the context of applying machine learning algorithms for deriving crime predictions (output). While complex machine learning models can optimize prediction performance, their opacity raises concerns about transparency and usability for decision-makers. Finally, this panel also examines the role of security technologies and different forecasting methodologies in improving crime prediction outputs and their practical implementation in law enforcement and public safety strategies (output).

This panel contributes to the ongoing discourse on improving the transparency, reliability, and effectiveness of spatiotemporal crime prediction models. By integrating empirical findings and methodological advancements, it offers insights for researchers, policymakers, and practitioners on the development and implementation of more robust and interpretable predictive models in crime prevention and public safety.

Subtopic

• Topic 8: Methodologies in Criminology / Advances in Quantitative Methods (Qantitative Methods WG)

Chair

• Wim Hardyns, Ghent University, Department of Criminology, Criminal Law and Social Law

Individual Presentations

• Formation processes of the criminological record - P. Jeffrey Brantingham, University of California Los Angeles

• Interpreting short-term burglary predictions: A comparative evaluation of XAI techniques - Robin Khalfa, Ghent University; Naomi Theinert, Ghent University; Wim Hardyns, Ghent University, Department of Criminology, Criminal Law and Social Law

• The performance-interpretability trade-off in crime prediction: Implications for decision-makers in public safety - Gaspard Tissandier, School of Criminal Justice Rutgers University-Newark; Alejandro Gimenez-Santana, Rutgers University

• Forecasting crime for strategic crime analysis: a comparison of methods - Matt Ashby, UCL

• Predicting Crime Hotspots: The Role of Security Technologies in Forecasting Residential Burglary in London - Alina Ristea, Assistant Professor Department of Security and Crime Science, University College London; Shane Johnson, Dawes Centre for future crime at UCL