Predicting Male Domestic Violence Using Explainable Ensemble Learning and Exploratory Data Analysis
This study addresses the under-recognized issue of male domestic violence in Bangladesh by employing exploratory data analysis and a high-performing, explainable stacking ensemble model (combining ANN, CatBoost, and Logistic Regression) to accurately predict risk factors and provide transparent insights for tailored interventions.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Domestic violence is a shadow that often falls on the most vulnerable, yet the shape of that shadow changes depending on who is looking. For decades, the conversation around abuse in the home has focused almost exclusively on women as victims and men as perpetrators, a narrative so dominant it has pushed other realities into the background. In many societies, including Bangladesh, cultural expectations of male strength and self-reliance create a silence around men who suffer abuse, leaving them without support or even a language to describe their pain. To understand this hidden landscape, researchers have turned to data, using the tools of modern computing to find patterns where human observation has failed. By gathering stories from thousands of individuals and applying statistical methods that can spot subtle connections, scientists can move beyond simple assumptions to see the complex web of factors that lead to abuse. This approach does not replace human empathy but offers a clearer map of the problem, revealing how economic stress, family structures, and relationship dynamics intertwine to create risk.
A team of researchers in Bangladesh set out to illuminate the specific experience of men facing domestic violence, a topic that has been largely overlooked in both academic study and public policy. They began by listening to 2,000 married men across nine major cities, asking detailed questions about their lives, their marriages, and their experiences with abuse. The data they collected told a difficult story: while the majority of men reported no abuse, a significant number did, yet these cases were often buried in a sea of non-abuse reports, making them hard to spot with traditional counting methods. The researchers found that the data was heavily skewed, with abuse cases appearing far less frequently than non-abuse cases, and filled with categorical details—like types of jobs, family structures, and locations—that are difficult for standard computer programs to process. To make sense of this messy, unbalanced information, they turned to a sophisticated form of machine learning, a type of artificial intelligence that learns by finding patterns in vast amounts of data.
The researchers tested a wide array of computer models, ranging from simple statistical tools to complex deep learning systems, to see which could best predict the likelihood of abuse. They found that the most successful approach was not a single model, but a team of models working together. They combined the strengths of different algorithms, creating a "stacking" ensemble where one model's predictions were fed into another to refine the answer. The most effective combination used a neural network, which mimics the way the human brain processes information, paired with a specialized model designed to handle categorical data, all overseen by a final decision-maker. This hybrid system achieved a remarkable level of accuracy, correctly identifying abuse cases 95% of the time and distinguishing them from non-cases with even greater precision. More importantly, the researchers did not just want a black box that gave an answer; they wanted to understand why the model made those decisions. By using explainable artificial intelligence techniques, they could peel back the layers of the computer's logic to see which factors mattered most.
The analysis revealed that the risk of domestic violence against men is not driven by a single cause but by a complex interplay of circumstances. The computer models identified that financial dependency plays a critical role; men who do not control their own income or who allocate a large portion of their earnings to their wives are at higher risk. Family structure emerged as another powerful factor, with men living in joint families—where multiple generations or relatives share a home—facing greater vulnerability than those in single-family households. The type of marriage also mattered, with arranged marriages showing different risk patterns compared to love marriages, and the duration of the marriage influencing the likelihood of abuse in non-linear ways. Perhaps most strikingly, the researchers found that factors like education level and profession, which might seem unrelated in a simple survey, became powerful predictors when viewed through the lens of the complex model. In a traditional statistical test, these factors might have appeared to have no connection to abuse, but the advanced machine learning revealed that their influence depends entirely on how they interact with other variables, such as income and family type.
The study challenges the prevailing notion that domestic abuse is a one-sided issue, providing concrete evidence that men are also victims and that their experiences are shaped by specific, identifiable social and economic pressures. The researchers demonstrated that while simple surveys might miss these nuances, advanced computational tools can uncover the hidden dynamics of abuse, offering a new way to understand the problem. They showed that the risk of abuse is not random but follows a pattern that can be understood and, potentially, addressed. By identifying high-risk groups—such as unemployed men, those in joint families, or those with specific financial arrangements—the study points toward the need for tailored interventions that go beyond general awareness. The work suggests that support systems and legal frameworks need to evolve to recognize these specific vulnerabilities, moving away from a one-size-fits-all approach to one that acknowledges the unique challenges faced by male victims.
Ultimately, this research serves as a bridge between raw data and human understanding, proving that technology can be used to give a voice to the silent. The findings do not offer a magic solution, but they provide a clear, evidence-based foundation for policymakers, social workers, and communities to build more effective support systems. By making the invisible visible, the study opens the door for a more honest conversation about domestic violence, ensuring that help reaches those who need it most, regardless of their gender. The path forward involves using these insights to design targeted programs that address financial stress, improve family dynamics, and create safe spaces for men to seek help, turning the cold numbers of a dataset into a warmer, more responsive reality for the people behind them.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.