Predicting Male Domestic Violence Using Explainable Ensemble Learning and Exploratory Data Analysis

Male domestic violence is significantly underrecognized: The study reveals that domestic violence against men in Bangladesh is a substantial problem that has been largely overlooked, challenging the common perception that domestic violence primarily affects women.

Financial dependency and verbal abuse are key risk factors: Exploratory data analysis identified that verbal abuse is highly prevalent in male domestic violence cases, and financial dependency plays a significant role in influencing these incidents, alongside familial and socio-economic factors.

Advanced machine learning can predict MDV with high accuracy: The researchers' stacking ensemble model achieved 95% accuracy and 99.29% AUC in predicting male domestic violence, demonstrating that data-driven approaches can effectively identify at-risk individuals despite data limitations and class imbalance challenges.

Explainability is essential for responsible AI in sensitive domains: The study used model-agnostic techniques like SHAP and LIME to make the predictive model's decision-making transparent and interpretable, which is critical for building trust and ensuring responsible deployment in sensitive social issues like domestic violence.

Data challenges require thoughtful methodological approaches: The research successfully addressed a 5:1 categorical imbalance and limited data availability through ensemble methods, demonstrating that responsible AI development must account for real-world data constraints rather than assuming ideal conditions.