Predicting obesity: a comparative analysis of machine learning models incorporating different features

Abstract

Obesity, the excessive accumulation of body fat, is a significant health risk associated with various detrimental impacts, including the development of chronic diseases, metabolic abnormalities, joint problems, sleep apnea, mental health issues, repro- ductive health difficulties, respiratory disorders, liver disease, and surgical risks. The emergence of machine learning, which offers potent analytical tools and high- performance computing capabilities, has revolutionised the interdisciplinary health industry. Through improved understanding and therapeutic interventions, this tech- nology offers opportunities to address and overcome the severe harm that obesity causes. This thesis aims to develop an automated system that utilises machine learning techniques to predict obesity based on different eating habits and relevant features. A comprehensive research methodology will be presented to categorise risk factors associated with an unhealthy lifestyle using machine learning. To effectively handle and anticipate various types of obesity, our AI system will analyse user data, including height, weight, daily food consumption habits, and more. The system will consider both weight-related and non-weight-related variables, as well as other fea- tures, to provide comprehensive insights into this health condition. Additionally, our technology will assist individuals by accurately classifying different forms of obesity, such as overweight I, overweight II, and beyond. Coefficient and correlation matri- ces have been utilised in the analysis to further enhance predictability. Therefore, by employing our obesity prediction algorithm, individuals can obtain estimates re- garding various levels of obesity. Empowered with this information, individuals can actively improve their health status by modifying their eating habits in accordance with their specific obesity condition. The primary objective of this research is to include and exclude features associated with predicting different levels of obesity and to see how this affects the accuracy scores. A secondary dataset and a range of machine learning techniques were employed to accomplish this goal, resulting in improved predictability and accuracy of the obesity-related outcomes.