Machine Learning Techniques for Neonatal Apnea Prediction

Shirwaikar, Rudresh D and Acharya, Dinesh U and Makkithaya, Krishnamoorthi and Surulivelrajan, Mallayasamy and Lewis, Leslie Edward Simon (2016) Machine Learning Techniques for Neonatal Apnea Prediction. Journal of Artificial Intelligence, 9 (1-3). ISSN 1994-5450

[img]
Preview
PDF
1323.pdf - Published Version

Download (179kB) | Preview

Abstract

Background and Objective: Machine learning has been widely accepted and applied in different fields to analyze data, but it is still novel in the field of neonatal diseases, especially neonatal apnea prediction. Apnea is a breathing problem associated with pathological changes in heart rate and oxygen saturation and is a common occurrence in neonates especially those who are born preterm. This study is focused on prediction of apnea episodes during the first week of childʼs birth using machine learning algorithms. Materials and Methods: The data consists of 229 examples of neonates admitted to Neonatal Intensive Care Unit (NICU) of Kasturba hospital, Manipal, Karnataka, India. This data is preprocessed and used to develop classification model using machine learning techniques such as decision tree (C5.0), Support Vector Machine (SVM) and ensemble approach, which includes random forest for prediction of apnea episodes. Results: The study compares models (decision tree, SVM and ensemble approach such as random forest) for accuracy. Among the results obtained, an accuracy of 0.88 and kappa of 0.72 using random forest algorithm for mtry three is found to be the most accurate model. Conclusion: The research work provides an automated machine learning based solution that helps clinicians predict apnea in neonates during the first week of their life. Inclusion of contextual information and preprocessing technique along with heterogeneous ensemble approach may further improve the models performance.

Item Type: Article
Uncontrolled Keywords: Machine learning, neonatal apnea, resampling techniques, support vector machine ensemble approach, bagging, boosting, random forest
Subjects: Engineering > MIT Manipal > Computer Science and Engineering
Depositing User: MIT Library
Date Deposited: 21 Oct 2016 08:53
Last Modified: 21 Oct 2016 09:03
URI: http://eprints.manipal.edu/id/eprint/147321

Actions (login required)

View Item View Item