Factors associated with the unmet need for spacing and limiting family planning: Evidence from a national survey

Traditional binary frameworks tend to combine the various aspects of unmet family planning needs into one general category. This study evaluates unmet family planning needs by first using conventional multinomial logistic regression to determine baseline demographic risk profiles. It then applies machine learning methods such as Random Forest and XGBoost to see if they can provide further insights into non-linear patterns and complex interactions between features that standard regression might miss. Using data from the Bangladesh Demographic and Health Survey (BDHS 2022), we built a multivariable multinomial logistic regression model. Additionally, we applied sophisticated ensemble classifiers such as Random Forest and XGBoost, paired with a synthetic hybrid resampling technique (SMOTE) to address severe class imbalance. Feature selection was conducted using the Boruta algorithm, and we used Shapley Additive Explanations (SHAP) to visualize complex global patterns and interaction effects. Bivariate analysis showed significant associations between unmet needs and maternal age, parity, wealth, residence, and decision-making power (p