A Retrospective Study on Obesity to Evaluate Omnipotence of Physical Condition Feature Set

Abstract

One of the growing medical concerns globally is obesity. The age old popular notion for the disease lies in physical conditions (PC), and eating habits (EH), leading to much observed debate for the root cause of obesity. This study establishes the omnipotence of PC over EH as a leading cause of obesity. The dataset used for the study comprised of 16 features which were divided into two feature subsets (FSS); one FSS containing 9 PC features, and the other FSS containing features related to EH. Initially obesity was classified using the complete feature dataset, followed by classification using the PC and EH FSSs respectively. Eight Machine Learning (ML) algorithms were used for the study. Regular performance metrics were used to evaluate the results. It was observed that the PC features unanimously contributed to obesity in contrast to EH features. Moreover, boosting was done using six algorithms, and results reflected that all the boosting algorithms enhanced the results. Of all the boosting algorithms, Hist-Gradient Boost generated the best results. The prime focus of the study is to analyze the major features for obesity using ML algorithms including boosting. This study computationally concludes that physical conditions have a greater impact on obesity with respect to eating habit conditions.