QUALITY-DRIVEN MACHINE LEARNING FOR NEONATAL CARE: PREDICTING NECROTIZING ENTEROCOLITIS

Abstract

Ensuring the quality and reliability of predictive models in neonatal healthcare is crucial for improving early disease detection and clinical decision-making. This study investigates the application of machine learning (ML) algorithms for predicting necrotizing enterocolitis (NEC) in neonatal populations, focusing on model selection, performance evaluation, and quality assessment. A dataset of 207 neonates, including 143 preterm and 64 term infants, was analyzed using six ML classification models: Logistic Regression (LR), Linear Discriminant Analysis (LDA), KNearest Neighbors (KNN), Classification and Regression Trees (CART), Naïve Bayes (NB), and Support Vector Machine (SVM). Model performance was assessed using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUROC). This study underscores the potential of machine learning in neonatal care and suggests that a hybrid approach combining highrecall and high-precision models could optimize NEC detection. Future research should focus on ensemble learning techniques and clinical validation to further enhance predictive performance and practical implementation in neonatal intensive care units.