Application of machine learning algorithms for prediction of abortion and its determinants among women of reproductive-age in East African countries

In low- and middle-income nations, abortion ranks among the top five causes of maternal mortality. It is associated to a pregnancy and childbirth-related complications. Despite this, there are few scientific researches focusing on predicting abortion and its determinants in East African countries. Therefore, this study aimed to predict abortion and its determinants among women of reproductive-age in East African countries. Community-based cross-sectional study design was used from eleven East African countries DHS dataset spanning 2015 to 2023. The study participants were all reproductive age women who were selected using a two-stage stratified sampling technique. The machine-learning algorithms were applied to predict abortion and its determinants using Python, particularly Jupiter notebook in Google colab. Data cleaning, one-hot encoding, data splitting, and ten-fold cross-validation were performed. Ten machine learning algorithms and SHAP were used to select and interpret the best model. From the total of 372,053 reproductive age women in East Africa, 12.2% participants perform abortion. Random forest was found the best model for training data with 91% of an AUC and 86% of accuracy. According to SHAPE analysis, women who have been never in union, women whose age 15-19 years, women whose age 20-24 years, women from Ethiopia, and women who not used any method were the top four features of abortion. This study identified that random forest classifier was emerged as the best-performing model to predict abortion among women reproductive age in East African countries. Marital status, marital status, age, country, Contraceptive use by method type, and living children plus current pregnancy were key determinant of abortion in East African countries. Governments and health systems should provide access to comprehensive family planning services, reproductive health education, and maternal health support.