Detecting Intimate Partner Femicide: A Machine Learning and Sensitivity Analysis Perspective

• In Event: Advances in Quantitative Methods Panel 1: Machine Learning and AI for Crime Analysis

Abstract

More than half of homicides against women are committed by their intimate partners or ex-partners. The high prevalence and severity of this violence have led to numerous studies focused on identifying associated factors and developing risk assessment instruments to prevent it. Despite these efforts, intimate partner femicide (IPF) continues to occur, highlighting the need for further research in this area. The current study explores new factors associated with IPF by applying Machine Learning (ML)-based clasification models -including logistic regression, tree-based models, instance-based learning, and neural networks-. This methodology enables the automatic detection of complex patterns in data, identifying factors related to both lethal and non-lethal violence against women within relationships among diverse cases. Additionally, a sensitivity analysis using SHapley Additive exPlanations (SHAP) was applied to the optimal ML model to identify the most relevant variables in the model’s decision-making process and quantify the impact of each variable’s contribution to the detection of types of violence in different cases. The findings evidence that ML models detect uncover common patterns among diverse cases, including less frequent patterns that are often overlooked in traditional statistical approaches. The most relevant factors identified are those related to the partner relationship and the environment, rather than the characteristics of the aggressor and the victim. The obtained knowledge has practical implications for guiding risk assessment strategies and professional interventions for the prevention of IPF, ultimately contributing to saving women's lives.

Author

• Esperanza Garcia-Vergara, Universidad Loyola Andalucia