Enhanced Prediction of Cardiac Risk in Neonates Using Calibrated Ensemble Learning Approaches

This study proposes a Heterogeneous Hybrid Machine Learning Model (HHMM) for the early detection of cardiac arrest in newborns in intensive care. The model combines three classifiers through stacking: XGBoost (XGB), Random Forest (RF), and Multilayer Perceptron (MLP), where their probability outputs serve as meta-features. The Random Forest classifier acts as a meta-learner, aggregating predictions for the final classification. The weighted stacking method assigns performance-based weights to the base classifiers by combining their probabilities using weighted averaging. The HHMM was trained and tested using clinical cardiovascular risk data. The model performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics. The results showed that HHMM outperformed the individual classifiers and conventional ensembles, achieving 95% accuracy, 96% precision, 94% recall, and 95% AUC-ROC. The HHMM's performance stems from its ability to capture complex patterns, reduce overfitting, and enhance generalization. The interpretability of the model through probability calibration makes it suitable for use in clinical decision-support systems. This study advances cardiac risk prediction in neonates using calibrated ensemble learning and stacking, thereby providing a foundation for real-time monitoring systems to improve patient outcomes in neonatal-care.