Predicting delayed antenatal care initiation among pregnant women in East Africa: using machine learning algorithms

Background Sub-Saharan Africa has the highest rate of maternal death due to pregnancy-related complications. The delayed onset of Antenatal care (ANC) is a major underlying factor for maternal mortality. The overall well-being and health of pregnant women and their fetuses greatly depend on the timely initiation of ANC care. This study aims to identify the top predictors of delayed antenatal care initiation in East Africa, including Ethiopia, to provide input for policymakers. Methodology We employed secondary data from the Demographic Health Surveys conducted across ten East African countries between 2016 and 2023. After preprocessing the data, which included cleaning and feature selection through Recursive Feature Elimination, we addressed class imbalance using Synthetic Minority Over-sampling Technique (SMOTE). We then applied seven supervised learning algorithms to develop a robust machine learning model. The LGBM Classifier emerged as the top performer for predicting delayed antenatal care initiation, achieving accuracy of 75%, an AUC score of 81%, an F1 score of 78%, a recall of 79%, and a precision of 77%. Additionally, we employed association rule mining to further analyze. Result Home delivery, low household income, a large number of household members, unemployment, being a younger woman, not using family planning methods, low educational level, and rural residence are predictors of delayed antenatal care initiation. Conclusion The prevalence of late antenatal care (ANC) initiation is high (65%). Promote family planning utilization, targeted health literacy campaigns, youth-friendly programs, expand mobile clinics, and promote skilled birth attendance to increase timely ANC initiation are recommended.