Using Machine Learning to Identify High Risk Domestic Violence Offenders in New York City

The researchers propose to use machine learning (ML) techniques to develop a tool that predicts risk of domestic violence (DV) victimization and to work with NYPD to test the efficacy of better targeting of high risk IPV victims through a large-scale randomized controlled trial.

The ML tool will combine administrative NYPD data with data from sources (e.g., 911 calls). The team will also use techniques to extract signals from text in officer reports, victim statements, and 911 calls.

Preliminary results suggest that an early version of the algorithm considerably outperforms both existing risk tools and business-as-usual decision-making by police officers. In partnership with NYPD, the team will formally test this via RCT of the tool to compare the effects of targeting officer home visits to victims—an intervention that has been found promising in quasi-experimental evaluations—using ML relative to status quo.

Key outcomes include rates of repeat victimization and serious injury, as measured by NYPD crime complaints and 911 calls (to capture calls for ambulance service, even if police are not called).

Note: This project contains a research and/or development component, as defined in applicable law,and complies with Part 200 Uniform Requirements - 2CFR 200.210(a)(14).
CA/NCF