Explainable machine learning algorithm to identify predictors of intention to use family planning among reproductive-age women in Ethiopia: Evidence from the performance monitoring and accountability (PMA) survey 2021 dataset

Abstract

Introduction: Approximately 225 million people in developing nations wish to delay or cease childbearing, but do not use any form of contraception. In the least developed countries, contraceptive usage was significantly lower, at 40%, and was particularly low in Africa at 33%. It is widely believed that intentions are a strong predictor of behavior, and many interventions that aim to change behavior including that targeting family planning use rely on evaluating program effectiveness through analyzing behavioral intentions. Understanding a woman's intention to use contraceptive methods is crucial in predicting and promoting the use of such methods. Therefore, this study aims to assess the determinants of intention to use family planning among reproductive age women in Ethiopia using explainable machine learning algorithm Method Secondary data from the cross-sectional household and female survey conducted by PMA Ethiopia in 2021 were used in the study. Using Python 3.10 version software, eight machine learning classifiers were used to predict and identify significant determinants of intention to use family planning on a weighted sample of 5993 women. Performance metrics were used to evaluate the classifiers. To smooth the data for additional analysis, data preparation techniques such as feature engineering, data splitting, handling missing values, addressing imbalanced categories, and outlier removal were used. Lastly, the greatest predictors of intention to utilize family planning were found using Shapley Additive exPlanations (SHAP) analysis, which further clarified the predictors' impact on the model's results. Result Using tenfold cross-validation and balanced training data, Random Forest revealed a performance of 77.0% accuracy and 85% areas under the curve, making it the most effective prediction model. The age at which family planning was first used, a partner or husband older than 40, being single, being Muslim, being pregnant, having previously been pregnant, needing to have more children, having a son or daughter relationship to the head of the household, and unmet needs for spacing and limiting were the top predictors of intention to use family planning, according to the SHAP analysis based on the random forest model. The research findings indicate that a range of personal and cultural factors may be taken into account when enacting health policies to enhance family planning intentions in Ethiopia. Therefore it’s highly recommended that the intention of family planning use and initiation of family planning provision should become a standard of service delivery to achieve the 2030 SDGs.