Photoplethysmography signals and physiological data in feature engineering and machine learning algorithms to calculate human-obesity-related indices

Abstract

The study developed a method based on photoplethysmography (PPG) and machine learning algorithms to predict three human-obesity-related indices: body mass index (BMI), visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT). This method eliminates the need for conventional, complex medical imaging examinations, such as computed tomography scans or magnetic resonance imaging. These conventional methods are not only time-consuming and expensive but computed tomography scans may also result in unnecessary radiation exposure to the body. PPG-based technology enables easy measurements without the need for complicated examination and measurement processes. In the proposed method, PPG signals are recorded and then processed to obtain statistical features, such as mean and variance. Subsequently, the measured data and extracted features are used in machine learning algorithms to predict human-obesity-related indices. Several feature engineering methods were employed to enhance the accuracy of our method, with the mean absolute errors for BMI, VAT, and SAT estimates decreasing from 0.419 to 0.228, from 0.624 to 0.563, and from 2.092 to 0.500, respectively. The results of the study indicate that combining PPG technology with machine learning and feature engineering methods is a convenient and effective method for measuring human-obesity-related indices. The information obtained through this method can enable individuals to understand their health status and adopt suitable measures for health management and disease prevention.