Identifying sub-groups of the obese from national health and nutritional status survey data using machine learning techniques

Abstract

The aim of this study is to discover unknown but meaningful patterns (groups) from the National Health and Nutritional Status and Survey (NHANSS) data by applying clustering analysis technique. This study focuses on determining the association of obesity, one of the leading causes of common non-communicable diseases, with the identified clusters. The subsets within NHANSS data were identified for obesity using various parameters including socio-demographic factors, physical activity and multiple dietary patterns of the participants. Due to the mixed data (qualitative and quantitative variables) in the dataset, two-step clustering technique with appropriate dissimilarity metric was chosen. Based on the twostep clustering method and association between the clusters and obesity factor, two clusters were formed and the anchoring characteristics of these clusters have been reported.