A Machine Learning and Hybrid Feature Selection Framework for Predicting Preterm Birth in Nulliparous Women Using Electronic Medical Records

Abstract

Preterm birth (PTB) is a leading cause of neonatal mortality worldwide, underscoring the need for accurate early prediction. This study presents a machine learning framework combined with hybrid feature selection to predict PTB risk in nulliparous women using data from electronic medical records collected during prenatal consultations. Clinical, demographic, and physiological data were obtained from the publicly available nuMoM2b dataset, covering three gestational intervals: 6–13 weeks, 16–21 weeks, and 22–29 weeks. Data preprocessing and standardization were performed using Differential Evolution (DE) to enhance quality and improve model performance. A hybrid feature selection approach, integrating the Dragonfly Optimizer with entropy-based relevance ranking, was employed to identify informative and non-redundant predictors, reducing dimensionality and noise while maintaining clinical interpretability. A Multilayer Perceptron (MLP) classifier trained on the selected features differentiated term and preterm deliveries. The framework achieved 88.41% accuracy, AUC = 0.80, precision = 85.84%, recall = 87.24%, and F1-score = 88.15%. Incorporating ultrasound features such as cervical length and Plasticity Index further improved predictive performance. Notably, at the third prenatal consultation, the model reached 85.62% sensitivity for predicting very preterm infants. These findings highlight the importance of ultrasound measurements and demonstrate that integrating machine learning with evolutionary optimization and entropy-based feature selection can significantly enhance early PTB risk detection. The approach enables timely interventions for high-risk pregnancies, potentially improving maternal and neonatal outcomes. This study underscores the value of computational methods in clinical decision support and emphasizes how machine learning can transform prenatal care for nulliparous women.