Automatic Diagnosis Metabolic Syndrome via a K−Nearest Neighbour Classifier

Behadada, Omar, Abi-Ayad, Meryem, Kontonatsios, Georgios and Trovati, Marcello (2017) Automatic Diagnosis Metabolic Syndrome via a K−Nearest Neighbour Classifier. GPC 2017: Green, Pervasive and Cloud Computing, May 11-14, 2017, Italy, pp. 627-637, ISBN 978-3-319-57186-7, ISSN 1611-3349, DOI https://doi.org/10.1007/978-3-319-57186-7_45.

[img]
Preview
Text
bdcat2016.pdf - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (936kB) | Preview

Abstract

In this paper, we investigate the automatic diagnosis of patients with metabolic syndrome, i.e., a common metabolic disorder and a risk factor for the development of cardiovasculardiseasesandtype2diabetes.Specifically,weemploy the K−Nearest neighbour (KNN) classifier, a supervised machine learning algorithm to learn to discriminate between patients with metabolic syndrome and healthy individuals. To aid accurate identification of the metabolic syndrome we extract different physiological parameters (age, BMI, level of glucose in the blood etc) that are subsequently used as features in the KNN classifier. For evaluation, we apply the proposed kNN algorithm against two baseline machine learning classifiers, namely Nave Bayes and an artificial Neural Network, on a manually curated dataset of 64 individuals. The results that we obtained demonstrate that the K−NN classifier improves upon the performance of the baseline methods and it can thus facilitate robust and automatic diagnosis of patients with metabolic syndrome. Finally, we perform feature analysis to determine potential significant correlations between different physiological parameters and the prevalence of the metabolic syndrome.

Item Type: Conference or Workshop Item (Proceedings)
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions: Computing and Information Systems
Date Deposited: 21 Mar 2018 16:02
URI: http://repository.edgehill.ac.uk/id/eprint/10177

Archive staff only

Item control page Item control page