A Machine Learning Approach to Prediction in Intimate Partner Violence Using Rival Models

Priya Devendran1

1Department of Sociology, University of Oxford


While research on criminal offending has utilised prediction to identify and target offenders for crime-control, such as greater levels of parole supervision, these applications have tended to rely on conventional statistical approaches to prediction that do not adequately account for the social realities of non-linear distributions. As a result, predictions have been fraught with problems, including high false positive rates. Novel algorithmic approaches have the potential to overcome this limitation, enabling superior prediction, although their application to intimate partner violence (IPV) is limited. To further understand how well non-linear distributions may be accounted for in IPV prediction, several different predictive approaches are examined to explore the strengths and weaknesses of each approach for predicting a sub sample of individuals who committed the most harmful IPV offences over a two year period.


Biography to come.