Intimate Partner Violence among Indian Ever Married Women: Exploring the Insights from Tree-Based Machine Learning Approaches

Abstract

Abstract Background: Intimate partner violence (IPV) is a global concern with severe consequences, encompassing physical, sexual, and psychological harm within relationships. Worldwide, 736 million women have experienced IPV, necessitating urgent attention. In India, despite a slight improvement, IPV remains prevalent, exacerbated by the COVID-19 pandemic.Addressing IPV requires understanding its determinants, rooted in societal norms and gender imbalances. This paper aims to bridge research gaps by applying machine learning alongside traditional methods to predict and understand IPV prevalence, considering different socio-economic and socio-demographic factors. Methods: The study has used data from the fifth round of National Family Health Survey. Descriptive statistics was used to analyse the sample characteristics.Multivariate logistic regression analysis was subsequently applied to determine the associations between IPV and associated risk factors. The instances of the prevalence of IPV was analysed using a combination of four distinct machine learning algorithms: decision trees (DTs), random forest (RF), gradient boosting (GB), and logistic regression (LR). Results:Prevalence of IPV among ever married women is found to be 68.71%.Older age, belonging to Scheduled Tribes, Other Backward Classes, being Hindu or Christian, employment status, higher number of children, marrying after 18, lower wealth quintile, husbands' alcohol consumption, regional disparities, rural residence, marital control, decision-making autonomy, justification of beating, and marital dynamics were all significant factors influencing IPV risk.Based on recall and F1 gradient boosting has better predictive performance than other machine learning models considered. The top ten predictors for IPV, included marital control, alcohol consumption, justified beating, region, decision-making autonomy, education years for both spouses, number of children, wealth index, and current working status. Conclusion: The study aimed to identify women vulnerable to IPV using three tree-based machine learning models on data from a national survey in India. The preference for gradient boosting was highlighted for its higher sensitivity, crucial for accurately identifying women genuinely at risk of IPV.Further the study encompassed the use of logistic regression as a base model for interpretation, revealing hidden patterns and relationships through machine learning analysis. Overall, the research contributes valuable insights into IPV among Indian women within the context of machine learning.