MACHINE LEARNING FOR ENHANCING EARLY MATERNAL AND FETAL HEALTH CARE - AN INTELLIGENT RISK ASSESSMENT FRAMEWORK

Abstract

Global public health still places great emphasis on maternal and fetal health care, particularly in underdeveloped nations where access to timely and high-quality prenatal services is sometimes constrained. Early detection of pregnancy-related hazards remains difficult despite developments in medical science because of a reliance on subjective clinical judgment and static threshold-based evaluations. Using machine learning (ML) technologies, this thesis offers an intelligent, data driven framework for the prediction and stratification of maternal and fetal health hazards. By means of early risk identification, the proposed system seeks to move from conventional reactive care models to proactive, tailored interventions. We examined a systematic maternal health database of 6,103 clinical records. Included were essential physiological and biochemical markers including systolic blood pressure, heart rate, glucose levels, HbA1c, body temperature, and body mass index. Fifteen ML models were run and assessed using accuracy, F1-score and AUC measures after thorough preprocessing including outlier management, multicollinearity reduction and feature scaling. With CatBoost reaching an accuracy of 98.61% and showing great interpretability using SHAP (SHapley Additive Explanations), ensemble models like CatBoost, XGBoost and LightGBM outperformed baseline classifiers. Addressing a significant drawback of current binary classification systems, the system classifies pregnancy risk into three categories: low, medium, and high. Furthermore, a fetal health classification module was created from CTG (cardiotocogram) data, so allowing complete prenatal evaluation. Both models were included into a Streamlit-based web application, therefore offering medical professionals a simple interface for real-time risk prediction and visual explanation of findings. By means of a scalable, interpretable and accessible clinical decision support tool, this thesis not only confirms the efficacy of ensemble ML models in maternal and fetal risk prediction but also stresses deployment readiness. Particularly in under-resourced areas, the solution is set to help doctors make educated, data-backed decisions that could greatly enhance maternal and fetal outcomes.