Predicting Short Time-to-Crime Guns - a Machine Learning Analysis of California Transaction Records (2010-2021)

Aug 5, 2026

Upload by Noah

Gun-related crime remains an urgent public health and safety challenge across the United States. A central question is how firearms are diverted from legal retail into offenders’ hands. Using nearly 8 million California legal firearm transaction records (2010–2020) linked to over 380,000 recovered crime gun records (2010–2021), the study applies supervised machine learning to predict which guns are used in crimes soon after purchase. Models use random forest with stratified under-sampling and SHAP-based interpretation, achieving strong discrimination (test AUC 0.85).

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Abstract

Introduction

Research gap and data constraints

Previous findings on risk factors

Study approach and aims

UC Davis UC Davis Previously Published Works Title Predicting Short Time-to-Crime Guns: a Machine Learning Analysis of California Transaction Records (2010-2021) . Permalink [https://escholarship.org/uc/item/7dp1j2pb](https://escholarship.org/uc/item/7dp1j2pb) Journal Journal of Urban Health, 101(5) Authors Laqueur, Hannah Smirniotis, Colette McCort, Christopher Publication Date 2024-10-01 DOI 10.1007/s11524-024-00909-0 Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California Journal of Urban Health (2024) 101:955–967 [https://doi.org/10.1007/s11524-024-00909-0](https://doi.org/10.1007/s11524-024-00909-0) ORIGINAL ARTICLE Predicting Short Time-to-Crime Guns: a Machine Learning Analysis of California Transaction Records (2010–2021) Hannah S. Laqueur · Colette Smirniotis · Christopher McCort Accepted: 20 June 2024 / Published online: 5 September 2024 © The Author(s) 2024 Abstract Gun-related crime continues to bean urgent public health and safety problem in cities across the US. A key question is: how are ﬁrearms diverted from the legal retail market into the hands of gun offenders? With close to 8 million legal ﬁrearm transaction records in California (2010–2020) linked to over 380,000 records of recovered crime guns (2010–2021), we employ supervised machine learning to predict which ﬁrearms are used in crimes shortly after purchase. Speciﬁcally, using random forest(RF)with stratiﬁed under-sampling, we predict any crime gun recovery within a year (0.2% of transactions) and violent crime gun recovery within a year (0.03% of transactions) . We also identify the purchaser, ﬁrearm, and dealer characteristics most predictive ofthis short timeto-crime gun recovery using SHapley Additive exPlanations and mean decrease in accuracy variable importance measures. Overall, our models show good discrimination, and we are able to identify ﬁrearms at extreme risk for diversion into criminal hands. The test set AUCis 0.85 for both models. For the model predict- Supplementary Information The online version contains supplementary material available at [https://doi.org/10.1007/](https://doi.org/10.1007/)[ ](https://doi.org/10.1007/)[s11524-024-00909-0](s11524-024-00909-0.)[.](s11524-024-00909-0.) H. S. Laqueur (B) · C. Smirniotis · C. McCort Violence Prevention Research Program, Department of Emergency Medicine, University of California, Davis, USA [e-mail: hslaqueur@ucdavis.edu](e-mail: hslaqueur@ucdavis.edu) H. S. Laqueur · C. Smirniotis · C. McCort California Firearm Violence Research Center, Davis, USA ing any recovery, a default threshold of 0.50 results ina sensitivity of 0.63 and a speciﬁcity of 0.88 . Among transactions identiﬁed as extremely risky, e.g., transactions with a score of 0.98 and above, 74%(35/47 in the test data) are recovered within a year. The most important predictive features include purchaserage and caliber size. This study suggests the potential utility of transaction records combined with machine learning to identify ﬁrearms at the highest risk for diversion and criminal use soon after purchase. Keywords Firearm transactions · Crime guns · Short time-to-crime · Random forest · Risk prediction · Variable importance Introduction Gun-related crime continues to be an urgent public health and safety problem in the United States. The ﬁrearm homicide rate increased by close to 35% in 2020 from the year prior [1] and rose another 8% in 2021, reaching a 29-year high [2] . In that year,ﬁrearms were used in 81% of the more than 20,000 homicides, the highest proportion reported in over 50 years [1]. This rise in homicides coincided with recordhigh ﬁrearm sales, which researchers have linked to increased gun violence [3, 4] . It also coincided with a signiﬁcant increase in ﬁrearms recovered in crimes shortly after legal purchase [5] . The rapid diversion of a ﬁrearm from sale to criminal use, i.e., a shor

FAQs

What problem does the study address?

The study examines how firearms diverted from legal retail are used in crimes shortly after purchase, i.e., short time-to-crime.

What data are used to build the machine learning models?

It links nearly 8 million legal firearm transaction records in California (2010–2020) with more than 380,000 records of recovered crime guns (2010–2021).

How well do the models perform and what key predictors emerge?

The test set AUC is 0.85 for both models, and important predictive features include purchaser age and firearm caliber size. Extremely risky transactions have a high recovery rate within a year.

Aug 5, 2026

Upload by Noah