Bio-impedance Analysis (BIA) Phase Angle Prediction with body competition data set using machine learning regression methods

Abstract

Purpose: Phase Angle (PhA) is a ratio of body reactance and resistance which can be determined by the means of bioelectrical impedance analysis (BIA). PhA is one of the most effective biomarkers that is being frequently applied in evaluating nutritional and cellular status. This study aimed to predict the PhA values in adults and determine the most significant factors that influence the PhA variations Methods: In this paper, we introduced a new approach based on Gaussian non-linear kernel support vector machine for the PhA prediction. The data was collected from 370 subjects (155 male and 215 female) age between 20-75 years by TANITA body composition analyser. Feature importance extraction was used to find the most relevant features to the PhA variations. The highest correlated features to predict the PhA value were introduced among the extracted twelve different features of body composition data along with age and sex. Results: A Gaussian support vector machine regression model was used to determine the most important features associated with PhA. Given the available data and performed machine learning model, it can be indicated that the PhA values, can be predicted with a reasonable degree of accuracy by measuring height, age, BMI, ECW (Extracellular Water), and ICW (Intracellular Water) values. Conclusions: Using the machine learning SVM regression method, we demonstrated the most important determinants of the PhA in adults’ bodies according to their body composition measurements. BMI, ECW (Extracellular Water), and ICW (Intracellular Water) values were determined as the significantly correlated features of the body composition data along with related individual’s age and height which together, highly affect the PhA variations.