Fetal Health State Classification, Based On Partograph Records Using Machine Learning Approaches: The Case Of Mizan Tepi University Teaching Hospital

Abstract

Fetal health is a significant public health concern, and monitoring it is crucial for mothers to ensure a healthy baby. In Ethiopia, a partograph is used during delivery to monitor labor conditions, maternal health, and fetal health presentation. Fetal and maternal mortality is a significant public health issue, particularly in low-income countries like Ethiopia, due to improper monitoring of fetal and maternal health during pregnancy and childbirth, resulting from inadequate record keeping. The researcher motivated because of in Ethiopia specifically the selected study area is experiencing an increase in maternal and fetal mortality due to complications from pregnancy and childbirth increasing in day to day’s activity. Pregnant women often die from underlying diseases, which are difficult to treat due to potential harm to the developing fetus. Continuous follow-up could fill this gap by providing accurate fetal state classification. The main objective of the study was to develop a classification model of the normal, suspect, and risky progress of feto-maternal and labor conditions based on a partograph presentation of recorded labor progress using machine learning approaches in the case of Mizan Tepi University teaching hospital. After collecting a partogram record of data required data preparation for analysis, data scaling to standardize the data, and preprocessing steps applied using the principal component analysis feature extraction method to handle big dimensional data keeping as much information in the data as possible. The classification was done with two scenarios with default hyperparameter and applying the model's respective hyperparameter tuning, such as CV and Gridsearch. Therefore support vector machine, random forest, and extreme gradient boosting classification machine learning algorithm were trained on the dataset and evaluated with classification evaluation metrics. Finally, the support vector machine scored higher on both scenarios and scored 99.19% after performance improvement was done with cross-validation and hyperparameter tuning. The developed models have the capability of categorizing fetal health status. It is recommended that collecting a large amount of data for all classes, and developing a more reliable classification model, and deployment in the future.