Predicting Bronchopulmonary Dysplasia in Infants: A Comparative Evaluation of Probit and Machine Learning Models

Abstract

This study compares the predictive performance of traditional Probit regression and several machine learning models in predicting Bronchopulmonary Dysplasia (BPD) among preterm infants. The models were evaluated using standard performance metrics, including accuracy, precision, specificity, sensitivity, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). Among all models, the Random Forest demonstrated superior predictive performance with the highest accuracy (86.36%), precision (85.71%), specificity (87.50%), sensitivity (85.71%), F1-score (0.8571), and AUC (0.92), indicating a strong discriminative ability. Birth weight and postnatal weight at four weeks emerged as the most significant predictors of BPD. The findings suggest that machine learning approaches, particularly the Random Forest algorithm, provide a more robust predictive framework than the conventional Probit regression model for early detection of BPD risk in preterm infants.