Predicting Risky Behavior Among Pregnant Women in Eastern African countries using Pooled Demographic and Health Survey (DHS) Data: An Explainable Machine Learning Approach

Abstract

Abstract Introduction Consumption of unhealthy substances such as tobacco use, and engagement in unsafe sexual activity are the major contributors to morbidity and mortality for pregnant women and the fetus. Although, risky behavior among pregnant women was complex and multifactorial public health concern, there is a limited study on those combined risk behaviors and their predictors among pregnant women in Eastern African countries using machine learning. Therefore, this study was conducted to examine the predictors of risky behaviors among pregnant women in Eastern African countries by using machine learning approach. Method The study used a cross-sectional study design from Demograpic and Health Survey data collected from 2012 - 2022 in Eastern African countries. STATA 17 was used for data extraction and preparation. Anaconda 3 and R softwares were used for data preprocessing, and analysis. Model performance was evaluated using accuracy, sensitivity, precision, F1-score and Area Under the Curve (AUC). Finally, the Shapley Additive Explanation (SHAP) was applied to further explain the predictors of risky behaviors among pregnant women. Results The pooled prevalence of risky behavior among pregnant women in Eastern African countries were 16.38% (95% CI: 12.63, 20.82). The Light Gradient Boosting Machine (LGBM) achieved an accuracy of 90% and AUC score of 0.96. The analysis revealed that pregnant women who lived in rural areas, being from poor and middle wealth income, women whose husbands had primary education, and women who do not exposed to media were among the most important predictors of risky behavior. Whereas women who are employed, women who utilized more than three Ante natal care service, and women aged between 25 and36 were contributed to a lower predicted likelihood of risky behaviors. Conclusion Risky behavior was a complex multifactorial public health challenge in Eastern African countries. The Light Gradient Boosting Machine (LGBM) was the best-performing model for predicting risky behaviors among pregnant women in Eastern African countries. The identified predictors provide valuable insights for developing targeted public health interventions to enhancing and increasing women’s access to more than three Ante natal care visits, enhancing husbands’ educational attainment, extending and implementing targeted media use, and empowering women economically. Expanding media campaigns and community outreach program to promote early and progressive adequate Ante natal care, particularly in rural areas.