International Journal of Science, Engineering and TechnologyAuthors: Logapriya A
Abstract: Gestational Diabetes Mellitus (GDM) is an increasingly serious maternal health concern, associated with high birth weight, premature delivery and a higher likelihood of Type-2 diabetes later in life. Conventional diagnosis relies on the Oral Glucose Tolerance Test (OGTT), which is usually performed late in pregnancy and needs laboratory infrastructure, delaying early intervention. Existing machine learning approaches, including XGBoost-based models, focus mainly on accuracy and depend heavily on preprocessing, while offering little in the way of risk categorization, visualization or patient record management. To address these limitations, this study presents a machine learning framework for early GDM prediction using the CatBoost classifier. The proposed system combines clinical and lifestyle parameters such as age, number of pregnancies, BMI, HDL, systolic and diastolic blood pressure, OGTT, hemoglobin, family history, PCOS, sedentary lifestyle and prediabetes status. Data preprocessing steps, namely missing-value imputation, removal of identifier and weak features, and standardization, are applied to improve data reliability. The trained model classifies patients as GDM or Non-GDM, estimates a probability-based risk percentage and assigns a risk category. A user-friendly Streamlit web application provides real-time prediction, a risk meter, an interactive dashboard, patient history storage and downloadable PDF reports. The model achieved an accuracy of approximately 96% on held-out data, with a confusion matrix showing only 40 misclassifications out of 3,525 records. Overall, the system offers an accurate, scalable and practical decision-support tool for early detection of gestational diabetes and improved maternal healthcare.
DOI:
