Prediction of visceral fat area of bioelectrical impedance based on ensemble learning

Abstract

Visceral fat as an important indicator of body composition largely reflects the health level of the human body. However, accurate measurement of visceral fat is still a big challenge. In this work, a new ensemble learning method named ByStepStack is proposed for visceral fat area prediction. The method is divided into two steps. In the first step, the mapping relationship model from the bioelectrical impedance to the body composition is obtained by Ridge regression. The second step is to predict the visceral fat area based on the body composition obtained from the previous step. The main innovation of the proposed method is using body composition to supervise feature representation, which can be a bridge between bioelectrical impedance and the visceral fat area. Finally, our method as well as the other state of the art ensemble learning methods is applied to predict visceral fat area on the same data set. The experimental results show that the proposed ByStepStack outperforms the other existing methods. Its relative error, average absolute error, mean square error and the determinable coefficients reach 0.0352, 2.237, 10. 7033, 0.9864 respectively.