Explainable Predictions of Preterm Birth: A LIME-Supported Ensemble Framework

Predicting preterm birth remains a critical medical challenge, as early identification can significantly reduce health risks for both mothers and newborns. In this study, we investigate the effectiveness of machine learning techniques for accurately predicting preterm birth. A comprehensive exploratory data analysis (EDA) was first conducted using correlation analysis, feature distribution visualizations, and pie charts to better understand and preprocess the dataset. After data normalization, several classifiers - including Random Forest, AdaBoost, Decision Tree, K-Nearest Neighbors (KNN), Logistic Regression, Naive Bayes, Neural Network, Support Vector Machine (SVM), and XGBoost - were applied for prediction. Among the individual models, KNN and Random Forest achieved the highest accuracy of 99%. An ensemble learning approach further improved the performance, reaching a maximum accuracy of 99.14%. To enhance model transparency and clinical interpretability, the Local Interpretable Model-Agnostic Explanations (LIME) technique was employed to explain individual predictions. The results demonstrate that the proposed approach can reliably predict preterm birth and has strong potential to assist healthcare professionals in early risk assessment and decision-making.