International Institute for Population Sciences
Deemed to be University (An Autonomous Organization of Ministry of Health & Family Welfare, Government of India)

Introduction and Objective: Declining fertility rates pose critical demographic and socioeconomic challenges. Identifying key determinants requires advanced methods capturing complex, non-linear relationships. This study uses machine learning with data imputation and feature selection to uncover major fertility drivers. Methodology: Using NFHS-5 data specific to Sikkim (n = 3,271; 45 variables), missing data were classified as MCAR, MAR, or MNAR through Chi-square tests and logistic regression. Imputation was performed using Mode, KNN, and MICE methods. Feature selection integrated filter (χ², Mutual Information, Pearson’s r), wrapper (RFE, Stepwise Regression), and embedded approaches (Random Forest, XGBoost, CatBoost, LightGBM, LASSO, SHAP, Permutation Importance). Models were trained with a 70:30 split and evaluated using MAE, RMSE, and R². Results: The study analysed NFHS-5 data from Sikkim (n = 3,271; 45 variables), addressing 18.9% missing values through systematic classification and imputation. Model performance improved significantly after imputation and feature reduction. Using all features post-imputation, XGBoost and CatBoost reached R² = 0.8029, while with the top 15 predictors, LightGBM achieved R² = 0.8561 and CatBoost R² = 0.8551 with RMSE reduced to 0.47. Overall, post-imputation improvements were statistically significant (ΔR² ≈ +0.10; p < 0.01), confirming the effectiveness of integrating imputation with feature selection in predicting fertility outcomes. Conclusion: Combining imputation strategies with feature selection enhances model accuracy and interpretability, uncovering robust fertility determinants. Findings highlight that delayed childbearing, rising education levels, contraceptive use, and fertility preferences are primary drivers of low fertility. These insights offer critical guidance for policy interventions, reproductive health programs, and sustainable demographic planning.