Comparison of Clustering Methods for Obesity Classification

Abstract

Body mass index (BMI) is mostly used as a reference through its indirect measure of fat mass and can be used conveniently. Despite such reference and convenience, in accordance to previous studies done, there exist a poor degree of agreement in obesity classification when it comes to BMI and the percent body fat that was found. Together with the utility of such obesity classification which refers to predefined cut-off values of BMI was seen as controversial. This study aims to discover a new method to classify obesity by using artificial intelligence (AI) techniques and statistical methods for obesity classification with minimum number of body dimensions required for input. The performance of methods used undergo comparison in terms of accuracy and interpretability. Results have shown that fuzzy rule-based system (FRBS) to be the most appropriate method amongst the rest. FRBS showed a performance of accuracy similar to other AI algorithms and discriminant analysis (DA), also showing a more stable and consistent provision of classification rules compared to the others. Concurrently, this study is suggesting the FRBS method as an obesity classification method.