# Machine learning algorithms for predicting and identifying the influencing predictors of antenatal care visits among women in Bangladesh: Evidence from BDHS 2022 data

## TL;DR

This study uses machine learning to predict antenatal care visits in Bangladesh and identifies key factors influencing them.

## Contribution

Applies ten machine learning models to BDHS 2022 data to predict and identify predictors of ANC visits in Bangladesh.

## Key findings

- The ranger (RG) model achieved the highest performance in predicting ANC visits with an accuracy of 69.46%.
- Age, wealth index, region, husband’s education, respondent education, and place of residence were identified as influential predictors.
- RG model results can guide targeted public health strategies to improve ANC utilization in Bangladesh.

## Abstract

Bangladesh, a South Asian country, continues to face significant challenges in maternal health, as reflected by its high maternal mortality ratio (MMR). According to the 2022 Bangladesh Demographic and Health Survey (BDHS), the MMR is 156 deaths per 100,000 births. This figure highlights ongoing challenges in maternal healthcare, despite improvements in recent years. Utilizing antenatal care (ANC) is a crucial intervention for reducing maternal mortality, as it enables early detection and treatment of complications, promotes health-seeking behavior, and prepares women for a safe childbirth. Thus, this study aimed to apply machine learning algorithms to predict the status of ANC visits and identify influential predictors among women in Bangladesh.

The study used BDHS 2022 data of 5,128 women aged 15–49 years. The outcome variable was ANC, defined as having at least four visits during pregnancy. We employed Boruta and Stepwise regression to identify the important predictors associated with ANC. Subsequently, ten different machine learning algorithms— decision tree, random forest, artificial neural network, logistic regression, adaptive boosting, extreme gradient boosting, gradient boosting, k-nearest neighbors, ranger (RG), and support vector machine—were trained on the training set to predict ANC visits. The predictive performance of the models was evaluated using accuracy, precision, recall, F1-score, and AUC on the test set,

The RG model performed best in predicting ANC visit status, with an accuracy of 69.46%, a precision of 68.51%, a Recall of 80.80%, an F1-score of 77.72%, and an AUC of 0.734, compared to the other models. The RG model identified age, wealth index, region, husband’s education, respondent education, and place of residence as the influential predictors of ANC utilization among women in Bangladesh,

The RG model and the identified influential predictors offer valuable insights for designing targeted public health strategies to enhance ANC utilization among women in Bangladesh.