Comparison of feature selection methods for estimation of body fat percentage

Obesity, which is one of the common health problems of our age, causes many discomforts as well as its negative impact on the quality of life of the person. Body fat percentage is the most important indicator in diagnosing obesity. Various devices and equipment are available for measuring body fat percentage. However, these are costly to measure and their use is limited. Determining this value quickly, easily, inexpensively and with high accuracy is as important as diagnosing obesity. It is possible to reliably calculate the body fat percentage value, which can be calculated from anthropometric data with machine learning algorithms. Body fat percentage, which is considered as a regression problem, has been successfully estimated by many studies in the literature. However, the presence of high-dimensional, irrelevant and redundant data in the data set distort the accuracy of machine learning algorithms and increases the training time of the model. There are feature selection algorithms that provide higher accuracy by using machine learning algorithms with fewer features. Thanks to feature selection, working with less data reduces the computational cost, as well as reducing the effect of the size problem, improving the prediction performance of the model, accelerating the learning process and providing a better understanding of the problem by machine learning models. In this study, eight different feature selection algorithms for body fat percentage estimation have been compared and higher accuracy results have been obtained with fewer features. Four machine learning methods have been used to examine the effect of feature selection methods on different models. The training times of these machine learning algorithms have been compared. As a result of experimental studies, it has been shown that by using feature selection methods, higher accuracy predictions can be obtained by spending less time on training the model with fewer features.