EUVIMEDEUVIMED entdecken
Crossref

Learning to predict violence: Deep learning approaches to violent crime prediction during pretrial release

Matthew DeMichele, Ian A. Silver, William Parkin

Criminology & Public Policy · 30.05.2026

DOI: 10.1111/1745-9133.70022 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Criminology & Public Policy, online erschienen am 30.05.2026. Verfasst von Matthew DeMichele und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 63 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Research summary This study investigates whether deep learning models—specifically feedforward neural networks—can enhance the prediction of new violent criminal arrests among individuals released pretrial. Using data from a large southeastern county, we evaluate four neural network configurations varying in depth, regularization, and class imbalance adjustments. Results show that incorporating class weighting and threshold tuning notably improves recall of rare violent events, with the best performing model identifying 57% of individuals arrested for violent crimes while maintaining stable false positive (FP) rates. We estimate counterfactual risk estimates among detained individuals, revealing that although some violent events may be averted through detention, this comes at the cost of detaining large numbers of FPs. Policy implications Improving the identification of individuals at risk for violent crime during pretrial release is a critical public safety priority, yet prevailing tools offer limited recall for such rare events. This study shows that neural networks, when properly calibrated, can improve detection of high‐risk individuals without increasing FP rates. However, algorithmic advances alone are insufficient; progress requires modernizing criminal justice IT infrastructures to enable interagency data linkages and real‐time analytics. Most current data systems lack the contextual and behavioral features necessary to capture latent risk. Ultimately, system improvement efforts should move away from mechanisms like monetary bail and toward evidence‐based supervision strategies, paired with investments in data modernization and cross‐agency collaboration to strengthen the predictive foundation of pretrial assessments.

Zitieren

APA

DeMichele, M., Silver, I. A., & Parkin, W. (2026). Learning to predict violence: Deep learning approaches to violent crime prediction during pretrial release. Criminology & Public Policy. https://doi.org/10.1111/1745-9133.70022

Vancouver

DeMichele M, Silver IA, Parkin W. Learning to predict violence: Deep learning approaches to violent crime prediction during pretrial release. Criminology & Public Policy. 2026. doi:10.1111/1745-9133.70022

BibTeX
@article{DeMichele2026,
  author = {DeMichele, Matthew and Silver, Ian A. and Parkin, William},
  title = {Learning to predict violence: Deep learning approaches to violent crime prediction during pretrial release},
  journal = {Criminology & Public Policy},
  year = {2026},
  doi = {10.1111/1745-9133.70022}
}
Kostenlos starten →