ARTIFICIAL INTELLIGENCE TO PREVENT VIOLENCE AGAINST WOMEN
Description
Violence against women constitutes a human rights violation and a persistent problem of public health, social justice, and effective access to protection. This article analyzes scientific and normative evidence on the use of artificial intelligence for prevention, with attention to predictive risk assessment, text-based detection, digital support, estimation of underreporting, and institutional governance. The review followed the PRISMA 2020 guidelines and covered publications released between January 2015 and June 2026. The initial search retrieved 1,512 records; after deduplication, screening, and full-text assessment, the corpus comprised 52 documents: 42 empirical studies, five reviews or protocols, and five normative frameworks. The results show greater maturity in clinical and police models. Several studies reported accuracy near or above 85%, although external validation, calibration, and fairness testing remain limited. Natural language processing achieved F1 values close to .80 across different samples, while multimodal models obtained the highest descriptive performance. Applications also included conversational assistants, referral to services, moderation of digital violence, and analysis of missing responses. The main risks concerned selection bias, data leakage, false negatives, false positives, opacity, and the absence of services after an alert. The article concludes that artificial intelligence can expand early identification and organize complex information, but its usefulness depends on professional oversight, local validation, data protection, independent audits, and effective pathways to care.
Files
Institutions
- Universidad de San Carlos de GuatemalaGuatemala, Guatemala City