Application of machine learning algorithm for prediction of abortion among reproductive age women in Ethiopia

Abortion is a critical health issue that leads to numerous complications, maternal deaths, and significant financial burdens on women, families, and healthcare systems. Studies have identified factors associated with abortion using traditional statistical analysis methods; however, no previous research has utilized machine learning to predict abortion in Ethiopia or identify its predictive factors. Machine learning is more effective and offers a better solution as it can capture complex and non-linear relationships in the data, leading to improved prediction accuracy compared to traditional regression models. Therefore, this study employed machine learning algorithms to predict abortion in Ethiopia and identify its predictors using nationally representative data. This study used the recent 2016 Ethiopian Demographic and Health Survey and included a sample of 14,931 women of reproductive age (15-49 years). This study used 7 machine learning algorithms for the classification of abortion. The dataset was randomly split into training and testing sets, with 80% allocated for training and 20% for testing. To evaluate the performance of each predictive model, we used a range of metrics such as accuracy, precision, recall, F1-score, and area under the curve (AUC). In this study, SHapley Additive Explanations (SHAP) values were used to measure the influence of each feature on the model's predictions. In the current study, 7 machine learning algorithm (i.e. logistic regression, decision tree classifier, random forest classifier, support vector machine, K neighbor classifier, XGBoost, and Nave bayes) were applied. The random forest classifier model were the best predictive models with the accuracy of 0.91 and AUC of 0.97. Moreover, the XGBoost was the 2nd best-performing algorithm with 0.87 accuracy and 0.94 AUC. According to the SHAP beeswarm and bar plots, younger age was identified as the strongest predictor of abortion, with a mean SHAP value of + 0.060. The second most impactful factor was having a younger husband, contributing a mean SHAP value of + 0.050 to abortion prediction in Ethiopia. Additionally, giving birth for the first time before the age of 18 ranked third, with a mean SHAP value of + 0.052. This study underscores the value of integrating machine learning into public health research and practice. Future work should focus on refining these models with larger and more diverse datasets, as well as exploring their applicability in other contexts and regions to further global maternal health initiatives. By harnessing machine learning techniques, healthcare providers can better classify abortion risks in reproductive-age women in Ethiopia. This knowledge can inform targeted interventions, enhance reproductive health services, and ultimately improve maternal health outcomes.