DEVELOPING A MACHINE LEARNING MODEL FOR CLASSIFICATION OF ANTENATAL CARE UTILIZATION

Abstract

Antenatal care is care given by a trained health care provider to a woman during pregnancy, starting from the time of conception up to the delivery of the newborn. Pregnancy complications are health problems that are caused by changes in physiological factors during pregnancy. Many pregnancy problems experienced by pregnant women can be prevented, detected, and treated during antenatal care visits. However, there are many pregnancy complications and maternal deaths in Ethiopia due to pregnancy-related causes. This indicates antenatal care use is still low in Ethiopia due to a lack of technology-related studies that can easily classify ANC utilization to support the ministry of health of Ethiopia in order to identify high-vulnerable areas of pregnancy problems and give special attention to them to reduce maternal morbidity and mortality. This study is aimed at identifying the factors that have an impact on the antenatal care follow-up of women during pregnancy and developing an antenatal care utilization classification model using machine learning techniques. To achieve the study's objective, the researchers used an experimental research design and a mixed research approach, both quantitative and qualitative. In this investigation, the Python tool on Google Colab, with its good features, was used to perform data preprocessing, data analysis, and model building. The eDraw Max tool was used to design the conceptual framework and proposed solution model. In this study, a total of 8929 instances with 17 attributes were collected from women who visited the antenatal and delivery care services in the hospitals. After data preprocessing, the researchers applied a domain expert suggestions and mutual information/information gain feature importance method to determine the important factors of antenatal care visits in order to classify antenatal care utilization. The randomized search CV algorithm was used to optimize the hyperparameters for each model. This study was conducted through experimentation using state-of-the-art machine learning algorithms such as MLP, DNN, DT, RF, and XGBoost classifiers to select outperforming models. The XGBoost classifier model conducted on all features based on domain expert outperforms others with an accuracy of 96.68%, a precision of 0.97, a recall of 0.97, an f1-measure of 0.97, and an AUC of 0.9979. Based on the experiment conducted, we conclude that the XGBoost classifier model was suitable to classify antenatal care utilization based on the factors that affect pregnant women during antenatal care visits