Article Text

Download PDFPDF
009 Machine learning analysis of firearm transaction records to predict risk of firearm suicide
  1. Hannah Laqueur,
  2. Colette Smirniotis,
  3. Christopher Mccort,
  4. Garen Wintemute
  1. UC Davis, Sacramento, USA


Background Firearms are by far the most lethal means of suicide, accounting for over half of suicide deaths. Evidence suggests that limiting access to firearms is an effective means of suicide prevention, yet accurately identifying those at risk to intervene and reduce access remains a key challenge.

Statement of Purpose To test whether California’s database of individual-level handgun transactions can be used to forecast firearm suicide risk and identify important individual and transaction-level predictors of firearm suicide.

Methods/Approach Using California’s Dealer Record of Sale database (1996 – 2015), we implement random forest classification to predict risk of firearm suicide within a year of purchase. We also estimate variable importance measures – i.e. each predictor’s contribution to forecasting accuracy.

Results Our cross-validated model AUC is 76. We identify individuals at increased risk: for example, individuals in the top 10% of predicted suicide risk accounted for close to half of all suicide deaths, with 90% specificity. Important predictor variables include known risk factors such as older age at first purchase as well as novel risk factors including firearm type (revolver) and proximity to the dealer.

Significance Our results suggest administrative data on firearm transactions can support efforts such as community-level suicide prevention partnerships between gun retailers and public health officials that focus on warning signs and the importance of means reduction.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.