Artificial Intelligence to fight against gender violence

A new model based on artificial intelligence improves predicting the risk of recurrence in gender violence.

The research, carried out by a team that included Juan Carlos Nuno from the Polytechnic University of Madrid (UPM) in Spain, aims to reduce the crime rate linked to gender violence.

Work on improving the assessment of the risk of recurrence in cases of gender violence based on data collected in Viogen, the Spanish national control and surveillance system for this type of crime, and applying this assessment to implement safeguards is concentrated. ,

The data collected in recent years in the VioGén system allow the application of machine learning (ML) techniques to assess the risk of recurrence of attackers and, consequently, to apply prevention measures according to the threat.

Machine learning uses computer programs with the ability to adapt to available data. These are algorithms that learn from experience (data) in a manner similar to an intelligent, albeit artificial, system. Learning allows the parameters of the program to be re-adjusted to make predictions or in this case risk assessments. “Unlike classical techniques based on static data, machine learning techniques allow for continuous updating as predictions are compared with actual data,” explains Juan Carlos Nuno.

25% more effective at assessing risk of misconduct

“Our study”, continues the UPM researcher, “proposes a better technique for assessing police security based on available resources: what is known as the nearest centroid (NC)”. This technique constitutes a powerful classification method that has been successfully applied in text classification and prediction of cancer classes from gene and expression profiles. In the context of this study, “the NC algorithm attempts to extract the salient features of each invasive archetype and, using them, analyzes new cases to calculate similarity with each of these common patterns,” Nuno says. .

In a certain sense, the UPM researchers point out, the operation of the NC model is analogous to some forensic methods, but “the large amount of data and the diversity of the responses allow machine learning techniques to extract very subtle information, information that is not there.” can be obtained directly by classical methods”.

As such, the proposed model predicts an improvement in risk assessment of 25% compared to the existing method implemented in Viogen. In addition, the (hybrid) model also includes a formula for applying the method currently present in the Viogen system and progressively integrates both methods according to the effectiveness index.

Along with Juan Carlos Nuno, researchers from the Complutense University of Madrid (UCM), the Secretary of State for Security and the Institute of Mathematical Sciences, all of Spain, are participating in the study. This work demonstrates that solutions based on automatic learning can be used effectively and accurately to predict the risk of recurrence in gendered violence, up to a marked improvement with respect to pre-existing prediction techniques. Can reach

The title of the study is “Hybrid Machine Learning Methods for Risk Assessment in Gender Based Crime”. And it has been published in the academic journal Knowledge-Based Systems.