Comparative Analysis of Classification of Neonatal Bilirubin by Using Various Machine Learning Approaches

Abstract

Background Neonatal jaundice poses significant risks to newborn health, necessitating early detection and management. Machine learning (ML) offers promising avenues for improving classification and monitoring, potentially revolutionizing neonatal care. Materials and methods A comparative analysis was conducted using various ML algorithms to classify neonatal bilirubin levels. Data were collected from neonatal images, and algorithms were trained and tested using standard methodologies. Performance metrics, including accuracy, precision, and recall, were evaluated to assess algorithm effectiveness. Results The Nu-Support Vector Classification (NuSVC) model emerged as the most effective, achieving a testing accuracy of 62.50%, with precision and recall rates of 61.90% and 56.52%, respectively. While variability existed among algorithms, these results highlight NuSVC's potential for clinical application in neonatal jaundice screening. Conclusion ML holds promise for improving neonatal jaundice detection and management. The findings suggest that the NuSVC algorithm can enhance screening accuracy, potentially mitigating risks associated with untreated neonatal jaundice. Future research should focus on refining models for broader clinical applicability and integrating ML into decision support systems to improve neonatal care globally.