Chinese study uses machine learning to unravel complex factors influencing breastfeeding duration

Researchers in China have developed an AI model that explores the intricate social and medical factors affecting how long mothers breastfeed, revealing interactions beyond traditional statistical approaches, with potential to tailor lactation support more effectively.

Researchers in China have used machine learning to tease out the mixed medical and social factors that shape how long mothers breastfeed, in a study published in PLOS One on August 7, 2026. The team said their goal was to move beyond standard statistical models, which can miss the way variables such as delivery method, milk stasis, education and feeding patterns interact with one another.

Using records from 210 postpartum women, the authors built an interpretable XGBoost model, a tree-based artificial intelligence system often used for tabular health data. According to the paper, the model was trained with polynomial feature engineering to capture higher-order interactions and SMOTE to rebalance the data set, then checked with SHAP, a method that shows which features push predictions towards higher or lower risk. The researchers reported an accuracy of 76.2% and a recall of 0.92 for women who breastfed for at least 11 months.

The most striking patterns involved combinations of factors rather than single variables. The study found that vaginal delivery appeared to soften the impact of prolonged milk stasis, while caesarean delivery seemed to intensify it. It also identified a negative interaction between education and feeding type, suggesting that women with higher education who shifted from exclusive to mixed feeding were at greater risk of stopping breastfeeding early. Those findings fit with earlier research showing that caesarean birth, maternal education and employment status all influence breastfeeding initiation and duration.

The authors argued that the model could support more targeted lactation care, particularly for women recovering from caesarean delivery and for those balancing breastfeeding with work-related pressures. The paper follows a growing line of research using machine learning to predict breastfeeding-related outcomes, including a 2026 Frontiers in Medicine study that identified early postpartum nursing factors linked to delayed lactogenesis after caesarean birth. Still, the PLOS One team said their work is limited by its single-centre design, modest sample size and reliance on self-reported milk stasis frequency, meaning larger prospective studies will be needed before the approach can guide routine care.

Disclaimer: This content is for informational purposes only and is not intended to be a substitute for professional medical judgment, advice, diagnosis, or treatment.