Predicting Domestic Abuse (Fairly) and Police Risk Assessment

Abstract

Domestic abuse victim risk assessment is crucial for providing victims with the correct level of support. However, it has been shown that the approach currently taken by most UK police forces, the Domestic Abuse, Stalking, and Honour Based Violence (DASH) risk assessment, is not identifying the most vulnerable victims. Instead, we tested several machine learning algorithms and propose a predictive model, using logistic regression with elastic net as the best performing, that incorporates information readily available in police databases, and census-area-level statistics. We used data from a large UK police force including 350,000 domestic abuse incidents. Our models made significant improvement upon the predictive capacity of DASH, both for intimate partner violence (IPV; AUC = .748) and other forms of domestic abuse (non-IPV; AUC = .763). The most influential variables in the model were of the categories criminal history and domestic abuse history, particularly time since the last incident. We show that the DASH questions contributed almost nothing to the predictive performance. We also provide an overview of model fairness performance for ethnic and socioeconomic subgroups of the data sample. Although there were disparities between ethnic and demographic subgroups, everyone benefited from the increased accuracy of model-based predictions when compared with officer risk predictions.

Keywords

• domestic abuse
• risk assessment
• machine learning
• algorithmic fairness
• police