A Thesis Submitted to the School of Graduate Studies in Partial Fulfillment of the Requirement for the Degree of Master of Science in Information Technology

Abstract

Accurate identification of child delivery modes, such as cesarean section (CS) and spontaneous vaginal delivery (SVD), is crucial for improving maternal and neonatal health outcomes. In regions like Arbaminch Town, Ethiopia, clinical assessments alone often lead to misclassifications in predicting delivery modes, resulting in unnecessary medical interventions, delayed CS procedures, and higher maternal complication rates. The growing application of machine learning (ML) in healthcare offers a promising solution, enabling data-driven decision support to optimize delivery mode predictions. The study bridges key research gaps in predicting child delivery mode by employing a comprehensive approach that combines advanced machine learning algorithms, deep learning algorithms, and temporal analysis that prior studies have not utilized all together. The general objective of this study is to develop a machine-learning model for predicting child delivery mode. The study focused on evaluating data availability, identifying key pregnancy factors, and selecting the best algorithm for predicting delivery mode. The study followed an experimental and exploratory research design with a mixed approach. The study utilized a dataset of 1,072 expectant mothers from the antenatal care report book of Arba Minch General Hospital and Birbir Health Center. The study employed a rigorous methodology involving data preprocessing, feature selection, missing value handling, and data standardization to prepare the dataset for model training. Six machine learning algorithms—Random Forest, Support Vector Machine (SVM), CatBoost, Gradient Boosting, Logistic Regression, and TabPFN, were trained and evaluated using cross-validation. Tools such as Python, Google Colab, and libraries like scikit-learn and TensorFlow were used for model implementation. The results showed that the Random Forest and TabPFN models achieved the highest accuracy of 93.1% and 92.5% respectively, with the 80/20 train test split. The CatBoost and the Gradient Boosting reached an accuracy of 91.2% and 91.1% respectively. The Random Forest model outperformed others in terms of precision, recall, and F1 score as wel