Automatic diagnosis metabolic syndrome via a k−nearest neighbour classifier

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 cardiovascular diseases and type 2 diabetes. Specifically, we employ the k-Nearest neighbour (k-NN) 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 k-NN classifier. For evaluation, we compare the proposed k-NN algorithm against two baseline machine learning classifiers, namely Naïve Bayes and an artificial Neural Network. Cross-validation experiments on a manually curated dataset of 64 individuals 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.