Predictors of fertility-limiting preferences among currently married women in Bangladesh: an interpretable machine learning analysis of the 2022 Demographic and Health Survey
Using interpretable machine learning on 2022 Bangladesh Demographic and Health Survey data, this study demonstrates that XGBoost can accurately predict fertility-limiting preferences among married women, identifying age, number of living children, and contraceptive intention as key predictors while revealing that digital inclusion and mental health have minimal influence.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the quiet calculus of family life, a woman's decision about whether to have another child is often the single most powerful predictor of what actually happens next. This choice, known as a fertility preference, acts as a compass for reproductive behavior. When a woman wants to stop having children, she is likely to seek methods to prevent pregnancy permanently or for a long time. When she wants to space births, she looks for temporary solutions. For decades, public health programs in countries like Bangladesh have relied on broad statistics to guess which women fall into which category, hoping to match the right services to the right people. However, as the total number of children per woman in Bangladesh has stabilized at a level that is still slightly above what is needed to simply replace the population, the question has shifted. It is no longer just about why fertility is high; it is about pinpointing exactly which women have already decided to stop childbearing so that health workers can offer them the long-term care they need, and which women are still building their families and need different support.
A team of researchers from Mawlana Bhashani Science and Technology University decided to tackle this problem using a modern approach called machine learning. Instead of relying on traditional statistical methods that assume every factor acts independently, they used computer algorithms capable of finding complex patterns in a massive dataset. They turned to the 2022 Bangladesh Demographic and Health Survey, a comprehensive national study that interviewed thousands of women. From this pool, they selected 17,245 currently married women between the ages of 15 and 49. The goal was simple yet ambitious: to see if a computer could accurately predict whether a woman wanted no more children based on a wide range of information about her life, including her age, her education, her mental health, her history of moving to a new place, and even her access to digital technology.
The researchers fed the computer twenty-five different pieces of information about each woman, such as how many children she already had, how long she had lived in her current town, whether she owned a smartphone, and if she had ever screened positive for anxiety or depression. They split the data into two groups, using the first group to teach the computer how to recognize patterns and the second group to test how well the computer performed on new, unseen cases. After training several different types of algorithms, they found that one model, known as XGBoost, was exceptionally good at the task. It correctly identified whether a woman wanted to stop having children about 84 percent of the time, a level of accuracy that rivals the best human experts and far exceeds what simple guessing could achieve.
The most striking discovery was not just that the computer could make these predictions, but exactly which factors drove those decisions. The algorithm learned that the strongest signals came from the most fundamental aspects of a woman's life stage. Her age and the number of living children she already had were the most powerful indicators. A woman in her late thirties with four or more children was almost certain to want to stop, while a young woman with no children was unlikely to. Beyond these obvious factors, the length of time a woman had lived in her current location proved to be a surprisingly strong predictor. Women who had lived in the same place for fifteen years or more were far more likely to want to stop having children, whereas women who had moved there recently were much more likely to want to have another child soon. This suggests that the stability of long-term residence is closely tied to the completion of family building, while recent movement often signals a family that is still in its early, growing stages.
The study also tested several ideas that had not been thoroughly explored in this context before, such as whether a woman's mental health or her access to the internet influenced her desire for more children. While the data showed that women with anxiety or depression, or those with high digital inclusion, did have different fertility preferences when looked at in isolation, the computer model revealed that these factors were not independent drivers. Once the model accounted for a woman's age, education, and number of children, the influence of mental health symptoms and digital access faded into the background. They were not the primary reasons a woman decided to stop or continue having children; rather, they were often just side effects of the life stage she was already in. Similarly, while wealth and geographic location mattered slightly, they were far less important than the personal history of the woman herself.
The implications of these findings are practical and immediate for family planning programs. Because the model can identify women who want to stop childbearing with high precision using only information that is routinely collected in surveys, health workers could use these same simple questions to target their services more effectively. They could focus on offering long-acting, permanent methods to women who have lived in a community for a long time and have several children, ensuring they do not face unintended pregnancies. At the same time, they could direct spacing services toward recent migrants who are likely still in the early phases of building their families. The research confirms that while the landscape of family life is complex, the decision to stop having children is not a mystery hidden in the noise of modern life; it is a clear signal written in the everyday details of a woman's age, her children, and her history of where she has lived.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.