Predictors of Community-Based Health Insurance Enrolment among Reproductive-Age Women in Ethiopia based on the EDHS 2019 Dataset: A Study Using SHAP Analysis Technique, 2024

Sisay Kassie ^1* Solomon Abuhay ¹ Mekdes Wondirad ¹ Samrawit Fantew ¹ Ayantu Melke ¹ Alex Ayenew Chereka ² Adamu Ambachew ² Abiy Tasew Dubale ² Yitayish Damtie ³ Habtamu Setegn Ngusie ⁴ Agmasie Damtew Walle ⁵

• ¹ Department of Public Health, College of Medicine and Health Sciences, Hawassa University, Hawassa, Ethiopia
• ² Department of Health Informatics, College of Health Sciences, Mattu University, Mattu, Ethiopia
• ³ Department of Public Health, College of Medicine and Health Science, Injibara University, Injibara, Amhara Region, Ethiopia
• ⁴ Department of Public Health, College of Health Sciences, Woldia University, Woldia, Ethiopia
• ⁵ Debre Birhan Health Science College, Debre Birhan, Ethiopia

Background: Out-of-pocket payment for health service leads to health catastrophes and reduces service utilization. To prevent this, community-based health insurance is an emerging strategy for providing financial protection against the cost of ill health. Despite the efforts made by the government of Ethiopia, the enrollment rate failed to reach the potential beneficiaries. Therefore, this study aimed to predict and identify predictors of community-based health insurance enrollment among reproductive-age women using SHapley Additive exPlanations analysis techniques.The study was conducted using the recent Demographic Health Survey 2019 dataset.Eight machine learning algorithm classifiers were employed on a total weighted sample of 9,013 reproductive-age women and evaluated using performance metrics to predict community-based health insurance enrollment using Python 3.12.2 version software with Anaconda extension. Furthermore, SHapley Additive exPlanation analysis was employed to identify the top predictors of community-based health insurance enrollment and the disproportionate effect of certain variables on another one.Result: Random forest was the best outperforming predictive model with a performance of 91.64% accuracy and 0.885% area under the curve. The SHapley Additive exPlanations analysis based on the outperformed random forest model revealed that residence, wealth, age of household head, husband educational level, media exposure, family size, and number of under five children were the top influential features that influences community-based health insurance enrollment.This study pinpoints the importance of machine learning for predicting communitybased health insurance enrollment and features hindering it. Residence, wealth status, and age of household head were found to be the top predictors. Findings from this study revealed that sociodemographic, sociocultural and economic factors might be considered while developing and implementing health policies intended to increase the enrollment of reproductive-age women in Ethiopia especially in rural areas of the country since it is a significant predictor that impacts low level of enrollment.