The responsible organisation

The Secretary of State for Security is a national Spanish Secretary that pertains to the Ministry of Interior. Under its umbrella are developed the government efforts to prevent and minimise gender and domestic violence, which are defined by the National Strategy for the Eradication of Violence against Women 2013-2016, the Strategic Plan for Equal Opportunities 2014-2016 and the State Pact against Gender Violence 2017-2022. The measures to fight against this type of violence are mandated by the national legislation, specifically in the Organic Law 1/2004, "for the Comprehensive Protection Measures against Gender Violence", and the Organic Law 3/2007, “for the effective equality among women and men”.

In this context, the web application “VioGén” has been implemented by the Secretary of State for Security of the Ministry of Interior and operates within the larger SARA network (Systems of Applications and Networks for Administrations), which is overseen by the Ministry of Economic Affairs and is the communications and services infrastructure connecting Spanish public administrations.

The problem

Violence against women and domestic violence are a dramatic persisting problem in the Spanish society. The secrecy of the events obstacles their identification and difficult effective prevention. Gender and domestic violence are often marked by fear, manipulation, and stigma, and victims may hesitate to report openly. Further, other specific complexities intensify the challenge. The probabilities of recidivism are high, as perpetrators tend to repeat and intensify the violent acts. 

Finally, in the Spanish public sector, many different entities work in the interventions and fragmented approach may obstacle effective risk assessment and timely intervention. For this, coordination and alignment among security bodies, judicial bodies and other professionals are essential to ensure the victims’ safety.

The solution and its implementation

The “Comprehensive Monitoring System in Cases of Gender Violence” (VioGén) is a probabilistic risk assessment tool used by the Spanish security forces and other entities involved in fighting against domestic gender violence. The system was launched in 2007 and updated in 2019 (VioGén 5.0). It is a pilot to improve the algorithm with AI was conducted in 2022. The algorithm is focused on strengthening gender-based violence recidivism prevention by estimating the risk that victims of domestic gender-violence face of suffering recidivism from the offenders. 

Specifically, it aims to achieve:

• Bringing together diverse organizations with authority in gender violence cases.
• Integrating all relevant data deemed necessary for effective risk assessment and protection measures.
• Utilising collected data to assess risks of re-offense.
• Based on the level of risk, providing protection and follow-up to victims throughout the national territory.
• Facilitating preventive work by issuing warnings, alerts, and alarms, through a subsystem of automated notifications, and guide targeted interventions to safeguard victims of reported cases.

VioGén takes the form of a web application to exchange information on the reported cases of domestic gender violence. Different levels of information can be accessed by authorised users from various entities, including:

• Forces and Corps of Security: national police and civil guard (mandatory), and regional and local police (voluntarily adhesion), 
• Penitentiary institutions, 
• Courts,
• Medical institutes legal and forensic sciences, 
• Victim assistance offices, 
• Prosecutor's offices, 
• Delegations and government sub-delegations,
• Social services and equality agencies of the different autonomous communities.

The algorithm estimates and predicts the risk victims face of suffering from recidivism in each reported case. The system employs a probabilistic risk assessment model composed of 37 variables. The factors assessed include the severity of past violent acts, the presence and nature of threats made by the perpetrator, the perpetrator's history of violence, and the degree of vulnerability of the woman and any children involved, among others. The system's algorithm analyses the data through predefined rules and risk assessment criteria to assign a specific risk level, which can range from "no perceived risk" up to "extreme risk." The level of police protection offered to the victim is defined based on this assigned risk category, ensuring a tailored approach to victim safety. 

The VioGén system uses different forms to collect the information needed to predict the risk of perpetrators’ recidivism:

• First, the Police Risk Assessment (“Valoración Policial del Riesgo” or VPR, in Spanish) form is used by police officers when a victim first reports an incident of gender-based violence. Through a series of standardised questions, the VPR categorises the risk levels of each reported case. Police officers are equipped with the “Zero Protocol”, which provides assistance during initial police interaction with victims, directing their efforts to accurately collect information needed to fill out the Victim's Profile.
• Second, the Police Evolution of the Risk Assessment (VPER or Valoración Policial de Evolución del Riesgo, in Spanish) form is used by the police, social services and law enforcement to monitor the situation and update the risk assessment on a regular basis. The forms are designed to be easy to use and understand, and they are available in multiple languages. 
• Finally, the system also includes the Forensic Risk Assessment form, or VFR for its Spanish denomination, which permits a deeper and more nuanced risk assessment, complementing the VPER information. It is used by forensic professionals, providing additional insight, and supporting a more nuanced evaluation of each case. Due to its exploration of victim and perpetrator profiles, the VFR requires a judicial order for its use.

Based on these forms, the system utilises the mentioned 37 coded risk indicators to calculate the risk faced by victims, whether after the initial report or subsequent ones. The risk assessments can also be modified and adjusted manually by officers if needed. Depending on the risk level, the protocol specifies the necessary police protection measures to enhance the victim's safety. These measures escalate in intensity as the risk level increases. For instance, in the case of “persistent” aggressors – those who have assaulted two or more women – the high-risk level of recidivism lead to additional protocol measures, including increased surveillance, monitoring of the aggressor and communication to the judicial authorities and victims themselves.

The role of police officers remains crucial in this type of risk assessment. Their experience in handling cases of violence against women enables them to interpret the context of the situation and adjust the risk assessment levels when needed, thereby integrating what the algorithmic system may overlook. Consequently, the system serves as a valuable supportive tool.

In 2022, after the 2019 release, a pilot experimentation has been conducted to utterly evolve the functionality of the system and the quality of the results, in particular with the experimentation of an evolved hybrid model that combines the statistical prediction methods (already used) with a more advanced Machine Learning (ML) method, permitting authorities to implement a smooth transition from the preexisting model to the ML-based model, with an improvement of up to a 25% with respect to the version 5.0 of the tool.

Expected benefits

From 2007 to 2023, more than 6 million cases were evaluated by the VioGén and more than 77 thousand are active cases. There are many benefits acknowledged by the entities using the system. 

• In the first stage, it promotes and facilitates coordinated actions between different institutions involved. 
• It offers a unified database on gender-based and domestic violence, reaching all the public entities. The different forms are integrated into a single protocol and enable the standardisation of the response along the national territory. 
• It guarantees a system of alarms that helps to keep track of cases with different levels of risk.
• It provides fundamental data and insights to advance scientific knowledge about gender-based violence and its underlying mechanisms.

Main benefits

While the VioGén system has proved to be a valuable tool for combating gender-based violence, certain challenges on its implementation and effectiveness persist.

First, obstacles related to the nature of the gender-based crimes have been identified. The definition of domestic and gender violence within the system might be limited, potentially excluding a significant portion of relevant cases. The evolving nature of these crimes presents a challenge, as the system needs to adapt to changing patterns and risk factors over time. 

Secondly, it has been observed that not all users have a similar and adequate training across the territory, as they pertain to different entities, security bodies, judicial bodies and other professionals. System users throughout different regions and entities may receive different levels of training, potentially impacting its effectiveness. To mitigate this, a manual has been developed as a centralised and unique source of information to use the system, complementing the training efforts made by different local, regional, and national forces and bodies. 

Third, improving the accuracy of the risk assessments has been a challenge as the assessment could drop false positives and false negatives, potentially under-protecting victims that may be in extreme danger. Additionally, there might be an automation bias, meaning that there is an over-reliance on the system’s automated risk levels, without manual adjustments performed by the officers when required. 

To enhance the system’s accuracy in assessing the risk that victims face, refinements of the gender-based violence predictors have been made, improving the forms and the algorithm.  As well, a new protocol was implemented in 2019, including the feature of identifying cases “of special relevance” where the risk of lethal violence is identified, and cases “with minors at risk”, where potential victimisation of children is observed. This enables the development of personalised protection measures for high-risk victims and their minor dependents. The refinement of the risk predictors was proven effective by recent studies, which show adequate accuracy parameters for the revised algorithm. 

• La valoración policial del riesgo de violencia contra la mujer pareja en España – Sistema VioGén
• The VioGén System exceeds six million risk assessments for victims of gender violence
• VioGén project 
• Machine Learning Applied to Gender Violence: A Systematic Mapping Study
• Hybrid machine learning methods for risk assessment in gender-based crime