Developing a Predictive Model for Gender-Based Violence in Urban Areas Using Open Data

Gender-based violence (GBV) in urban contexts is a complex, multifactorial phenomenon shaped by socioeconomic, territorial, and contextual factors. This study aims to develop a predictive model for GBV-related crimes in Valencia (Spain), using open geospatial data and advanced machine learning techniques to support the identification of high-risk areas and guide targeted interventions. A 25 m grid was generated to homogenize crime data and independent variables, including socioeconomic indicators, urban services, real estate information, and traffic intensity. Multiple models were tested—Multiple Linear Regression (MLR), Decision Tree (DT), and Random Forest (RF). Linear models were found to be insufficient for explaining GBV patterns (R2 ≈ 0.45), while RF and DT achieved high predictive accuracy (R2 ≈ 0.97 and 0.95, respectively. The variables with the greatest influence were traffic intensity, average monthly income, unemployment rate, and proximity to nightlife venues. To enhance the interpretability of the most accurate models, we applied SHAP (SHapley Additive exPlanations) to quantify the contribution of each predictor and elucidate the direction and magnitude of their effects on model predictions. These findings demonstrate the utility of geospatial ML techniques in understanding the spatial dynamics of GBV and in supporting urban safety policies. While the current model focuses on static spatial predictors and does not explicitly model temporal dynamics or spatial autocorrelation, future research will integrate these aspects, along with participatory data, and test the model’s applicability in other cities to enhance its robustness and generalizability.