Improving Neonatal Care with AI: Class Weight Optimization for Respiratory Distress Syndrome Prediction in Very Low Birth Weight Infants.

In this study, we developed an AI model to predict Respiratory Distress Syndrome (RDS) in premature infants, aiming to reduce unnecessary treatment with artificial pulmonary surfactant. We analyzed data from 13,120 infants in 76 hospitals, considering various factors including infant information, maternity details, birth process, family background, resuscitation, and lab results. seven machine learning algorithms were compared, with Support Vector Machine (SVM) showing the highest accuracy. We further improved prediction performance with a 5-layer Deep Neural Network (DNN) using selected features from SVM-based analysis. To address imbalanced data, we employed ensemble methods and class weight optimization. The final model achieved exceptional results on an independent test dataset, with a specificity of 87.36%, sensitivity of 90.65%, balanced accuracy of 89.01%, and an AUC of 0.9612, surpassing other models.