Özet:

Obesity is a significant health problem, and its prevalence is increasing. It is known that this disease is the trigger and precursor of many other diseases. Before the treatment of obesity disease, it is important to determine the body fat percentage correctly. Body fat percentage can be measured precisely with high-cost methods. However, these methods are not common to use. In this study, support vector regression and random forest tree regression methods were applied to accurately and cost-effectively estimate body fat percentage with anthropometric data set taken from individuals. In regression methods, model parameter values, number of data, number of features and feature selection are important in prediction performance. In the study based on a 13-featured body fat percentage data set, a new data set was created with 25 statistical methods (skewness, central moment, kurtosis, etc.) frequently used in the literature, and it was observed that the performance of the new data set was higher than other studies in the literature. Estimation accuracy has been increased by determining the regression parameters with grid scanning methods. In addition, features that are highly correlated with body fat percentage were determined with feature reduction methods. It was observed that the prediction success performance of the regression methods performed with the selected features was higher than other similar studies. As the best mean square error values, the value of 2.2519 was obtained in the experiment performed with the Random Forest Trees Method and the new data set created by the statistical method, while the value of 3.174 was reached in the regression experiment with Decision Support Machines and the properties with the best 6 f-score values.

Anahtar Kelimeler: