Abstract

Reducing newborn mortality by 2030 is a Sustainable Development Goals target 3.2. Neonatal sepsis is the third major cause of neonatal death after prematurity and birth asphyxia. Late-onset neonatal sepsis (LOS) refers to sepsis in neonates between the ages of 3 and 28 days and is likely to be acquired from the environment rather than maternal transmission. Since clinical features are not evident in the initial stages of infection, early diagnosis of LOS is challenging. Studies have shown that physiological parameters can predict LOS before prominent clinical features. Clinicians can use these parameters as early warning signs to monitor neonates closely and intervene earlier to prevent complications and provide effective treatment. This paper compares various machine learning algorithms to predict the onset of neonatal sepsis using vital signs, laboratory measurements, and observations captured within 24 hours of admission from the MIMIC III dataset. Experimental results show that adaptive boosting, light gradient boosting and random forest with Synthetic Minority Oversampling Technique give the highest area under the receiver operating characteristic (AUROC) of 0.9248, 0.9245, and 0.9238, respectively, among all the algorithms evaluated using 10-fold stratified cross-validation. The soft voting classifier trained on an ensemble of the top three models predicted the onset of neonatal sepsis with an AUROC of 0.9266, accuracy of 0.8553, F1 score of 0.7829, and Matthew's correlation coefficient of 0.6995.

Original languageEnglish
Article number976
JournalEngineered Science
Volume26
DOIs
Publication statusPublished - 2023

All Science Journal Classification (ASJC) codes

  • Chemistry (miscellaneous)
  • General Materials Science
  • Energy Engineering and Power Technology
  • General Engineering
  • Physical and Theoretical Chemistry
  • Artificial Intelligence
  • Applied Mathematics

Fingerprint

Dive into the research topics of 'Early Detection of Late Onset Neonatal Sepsis Using Machine Learning Algorithms'. Together they form a unique fingerprint.

Cite this