Classification of Obesity Using Several Machine Learning Techniques

Abstract

Several potential dangers are linked to obesity. It's a major reason why so many people are becoming sick and dying from chronic diseases. There are several obstacles in the way of uncovering the causes and effects of obesity. The standard regression method assumes independence and linearity among variables and restricts the number of predictors that may be examined. If you're looking for an alternative to traditional approaches to data analysis on obesity, consider using Machine Learning (ML) techniques. Using an innovative strategy with sophisticated machine learning techniques for forecasting obesity as an attempt to go beyond traditional prediction models, this study aims to assess the ability of three different ML methods to identify the presence of obesity using freely accessible health data. These methods are Logistic Regression, Classification and Regression Trees (CART), and Nave Bayes. Meanwhile, the primary purpose of this research is to identify, from among the available variables, a collection of risk factors for adult obesity. In addition, we use the Synthetic Minority Oversampling Technique (SMOTE) to predict obesity status from the known risk variables in order to resolve data imbalance. Based on the results of this analysis, Logistic Regression seems to be the most effective technique. However, the kappa coefficients reveal a weak agreement between the projected and actual rates of obesity. Adult obesity can be predicted by a number of factors, including geographical location, marital status, age range, level of education, consumption of sugary drinks, fat content/oily foods, grilled foods, food that has been preserved, seasoning powders, soft/carbonated drinks, alcohol, medically diagnosed hypertension, mental/emotional disorders, lack of physical activity, smoking, and consumption of fruits and vegetables. Health officials might use this information to better manage chronic illnesses, particularly those linked to obesity, if they knew what risk factors to look out for. Furthermore, employing ML approaches on publically accessible health data, such as Indonesian Basic Health Research (RISKESDAS), is a viable way to bridge the gap for a more solid understanding of the correlations between numerous risk variables in predicting health outcomes.