Machine Learning Approaches for Prediction of Fertility Determinants in Bangladesh: evidence from the BDHS 2017-18 data

Abstract Background Fertility is a social indicator that represents the country’s growth and economic sustainability. The fertility rate of a country refers to number of average children born to a woman during her lifetime. It is an important demographic indicator that influences population dynamics, economic growth, social welfare, and public policy. This research leverages advanced machine learning methodologies to achieve more precise predictions of fertility and fertility determinants in Bangladesh. Methods The dataset utilized in this study was sourced from the Bangladesh Demographic Health Survey (BDHS) conducted in the year 2017–18. Python 3.0 programming language were used to implement and test the machine learning (ML) models such as Random Forests (RF), Decision Tree (DT), K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), XGBoost, LightGBM and Neural Network (NN). We have used Boruta algorithm of Feature selection with R programming language packages. Conventional methods were analyzed using SPSS Version 25 and R programming language. The predictive models performance was evaluated and compared with the metrics such as macro average and weighted average of the Confusion Matrix, Accuracy, F1 Score, Precision, Recall, Area Under the Receiver Operating Characteristics Curve (AUROC) and K-fold cross-validation. Results We preferred with the Support Vector Machine (SVM) model of fertility in Bangladesh with macro average recall (93%), precision (89%), F1 score (90%) in addition with weighted average recall (97%), precision (96%), F1 score (96%) K-fold accuracy (95.9%). Our predictive models showed that Access to mass media, Husband/partner's education level, Highest educational level, Number of household members, Body Mass Index of mother, Number of living children and Son or daughter died stand out as the key determinants influencing fertility in Bangladesh. Conclusions In the realm of constructing advanced predictive models, Machine Learning methods surpass conventional statistical approaches in classifying concealed information. In our Study the Support Vector Machine (SVM) emerged as the top-performing model for fertility prediction in Bangladesh.