Revisiting the Factors Affecting Child Sex Preference among Married Women in Bangladesh using Machine Learning Techniques

Abstract Background: Sex preference for children remains an essential sociocultural determinant of fertility behavior in Bangladesh and many other countries in the world. The best-fitting model is crucial in determining the factors that affect sex preference. Objective: The research employs Logistic Regression, Random Forest, Support Vector Machine, Extreme Gradient Boosting, and Naïve Bayes to explore the factors influencing child sex preference. Methods: A cross-sectional study was conducted among 625 randomly selected married women in Dhaka city. Hyperparameter tuning, 10k-fold cross-validation, and dataset balancing using SMOTETomek were applied to enhance model performance. Results: The study explores which factors influence sex preference and identifying key predictors using the Support Vector Machine (SVM), achieved an accuracy of 80.0% with a more stable and interpretable outcome. To enhance interpretability, SHapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were utilized, identifying women’s empowerment, the number of children ever born, and the educational level of both the respondent and her husband as the most influential predictors of sex preference. Other variables such as contraceptive use, sex composition of living children, working status, and wealth index are notable predictors of sex preference. Factors such as religion, family type, and age at marriage show a relatively lower impact on sex preference but are still relevant. Women who experienced domestic violence showed some effect but may not have the most significant pull factor on preference for sex. Conclusion: By examining the factors that influence children's sex preferences, this study offers valuable insights into the socio-economic dynamics of gender preference, providing actionable guidance to promote gender-equitable attitudes in Bangladesh.