Utilizing Machine Learning to Predict Protective Factors for Childhood Exposure to Violence
Iris Chen
03/09/2026
Childhood exposure to violence is defined as the victimization or witnessing of violence by a child under 18 years of age, and affects over 60% of children in the United States. Though children exposed to violence consistently display more negative behaviors and outcomes, research has identified certain protective factors that mitigate negative effects and increase the likelihood of resilience. This study aims to investigate the relationship between various protective factors (family support, community support, community involvement, mentorship) and violence-impacted outcomes (behavior, self-motivation, academic performance, and emotional and developmental well-being) in a nationally representative population of children with exposure to violence. Utilizing data from the National Survey of Children’s Health, we examined the significance and relationship directionality of all four key protective factors in each outcome. Results were compared across three predictive models – logistic regression, decision tree, and Random Forest – revealing no substantive differences between the odds ratios and the feature importances of the non-parametric models. Family support emerged as the strongest predictor of positive outcomes across all three algorithms, with community involvement also demonstrating consistent importance. Additionally, factors such as bullying victimization and overall health status were identified as highly significant, despite not being included in the original set of key independent variables. Findings suggest that family-centered support and community engagement are particularly effective in buffering negative consequences of childhood exposure to violence. These results shed light on the importance of interventions that prioritize family-based programming, anti-bullying initiatives, and structured extracurricular opportunities to strengthen resilience and improve outcomes.