Machine learning-based fetal health prediction and development of smart web application.

Introduction: Fetal health monitoring is critical for early identification of pregnancy-related risks. Manual interpretation of cardiotocography (CTG) signals is subjective and variable among healthcare professionals. Methods: A machine learning-based framework was developed to classify fetal health into Normal, Suspect, and Pathological categories using CTG-derived clinical features. The dataset was preprocessed through duplicate removal, normalization, class balancing using SMOTEENN, multicollinearity analysis via VIF, and Kruskal-Wallis statistical feature selection. Eleven machine learning and neural network models were trained and compared, including Logistic Regression, K-Nearest Neighbors, SVM, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, XGBoost, LightGBM, Multi-Layer Perceptron, and Deep Neural Network. Results: LightGBM achieved the best overall performance with 96.03% accuracy, 91.99% balanced accuracy, 93.05% macro F1-score, 99.02% ROC-AUC, 88.91% Cohen's Kappa, and 89.01% MCC. SHAP-based explainability identified abnormal short-term variability and fetal heart rate accelerations as the most important features. Discussion: The best-performing LightGBM model was integrated into a Streamlit-based web application for real-time fetal health prediction, demonstrating its potential as a clinical decision-support tool.