RESUMO
BACKGROUND AND OBJECTIVE: Muscle mass is one of the critical components that ensure muscle function. Loss of muscle mass at every stage of life can cause many adverse effects. Sarcopenia, which can occur in different age groups and is characterized by a decrease in muscle mass, is a critical syndrome that affects the quality of life of individuals. Aging, a universal process, can also cause loss of muscle mass. It is essential to monitor and measure muscle mass, which should be sufficient to maintain optimal health. Having various disadvantages with the ordinary methods used to estimate muscle mass increases the need for the new high technology methods. This study aims to develop a low-cost and trustworthy Body Muscle Percentage calculation model based on artificial intelligence algorithms and biomedical signals. METHODS: For the study, 327 photoplethysmography signals of the subject were used. First, the photoplethysmography signals were filtered, and sub-frequency bands were obtained. A quantity of 125 time-domain features, 25 from each signal, have been extracted. Additionally, it has reached 130 features in demographic features added to the model. To enhance the performance, the spearman feature selection algorithm was used. Decision trees, Support Vector Machines, Ensemble Decision Trees, and Hybrid machine learning algorithms (the combination of three methods) were used as machine learning algorithms. RESULTS: The recommended Body Muscle Percentage estimation model have the perfomance values for all individuals R=0.95, for males R=0.90 and for females R=0.90 in this study. CONCLUSION: Regarding the study results, it is thought that photoplethysmography-based models can be used to predict body muscle percentage.