Functional Improvement of Neonatal Mortality Using Stack Ensemble Approach

Abstract

Globally, mortality of neonates remains one of the substantial outcomes to predict in Neonatal Intensive Care Units, where early and accurate risk prediction is crucial for timely medical intervention due to the exponential increase in neonatal deaths. The study aims to facilitate the functional improvement of neonatal mortality using stack ensemble technique with base classifiers, including Random Forest, Gradient Boosting, Support Vector Machine (SVM), and XGBoost, with a meta-classifier based on Logistic Regression to enhance accuracy and robustness of the prediction. Synthetic generated dataset utilized in this study consists of features such as birth weight, gestational age, preterm birth status, and presence of medical conditions. The dataset was pre-processed to ensure it was clean and fit for the machine learning algorithm. Using a constructive research methodology, this study employs an object oriented design approach, python-based frameworks and visualization tools to enhance interpretability and transparency in decision-making processes. This research demonstrates the power of ensemble learning in predictive analytics and providing risk evaluation, and actionable insights. Results demonstrated that all models achieved high predictive accuracy, with the stacking ensemble model outperforming individual classifiers in terms of sensitivity, specificity, and F1-score. The stacking model achieved an accuracy of 93%, precision of 94%, sensitivity of 96%, specificity of 84%, F1-score 88% and an AUC of 97%. The study demonstrates that ensemble model approach significantly enhances the effectiveness of neonatal mortality prediction compared to standalone models and would efficiently guide medical professionals on allocating adequate resources and interventions promptly in the intensive care units.