Decision tree-based statistical learning and quantile regression adjustment: Insights from pregnant women in Benin

Abstract

This research investigated factors influencing pregnancy outcomes (stillbirth or live birth) and newborn weight distribution among pregnant women in Benin. It covered 1,048 pregnant women who attended the five referral hospitals in Couffo in 2020, with data on 952 live births. We used random forests for feature selection, CART decision trees for pregnancy outcome prediction, and quantile regression to examine newborn weight distribution. The CART model identified key risk factors for adverse pregnancy outcomes, including young maternal age (18–24 years), a history of HIV, two or more prior stillbirths, and multiparity (four or more living children). In contrast, favorable pregnancy outcomes were associated with maternal age (25–40 years), at least three prenatal consultations, and the absence of major medical conditions. The neonatal death rate was 91.6 per 1000 live births. Quantile regression analysis revealed that low birth weight was associated with young maternal age (14–24 years), strenuous maternal occupation, paternal occupation, pregnancy complications (hypertension, preeclampsia, anemia, premature rupture of membranes), twin pregnancy, late prenatal care (third trimester), fewer than four prenatal visits, and a history of cesarean section. Macrosomia was linked to maternal age (25–45 years), parental occupation, and the absence of medical or pregnancy-related complications. Notably, macrosomia was strongly associated with a high cesarean section rate (64.44%). These findings highlight critical factors influencing pregnancy risks and newborn health, emphasizing the need for targeted prenatal care strategies in Benin.