XGBoost Regression of the Most Significant Photoplethysmogram Features for Assessing Vascular Aging

Abstract

The purpose of this study was to confirm the potential XGBoost as a vascular aging assessment model based on photoplethysmogram (PPG) features suggested in previous studies, and explore key PPG for through an explainable artificial intelligence method. waveforms obtained from 752 volunteers aged 19-87 years were analyzed total 78 derived that proposed studies. Age estimated regression model, estimation error calculated terms mean absolute root-mean-squared error. To evaluate feature importance, gain, coverage, weight, SHAP value calculated. developed using has 8.1 mean-absolute 9.9 error, correlation coefficient 0.63 with actual age, determination 0.39. Feature importance analysis confirmed features, such systolic diastolic peak amplitude, risetime, skewness, pulse area, play role assessment. showed equal level performance existing PPG-based models. Moreover, result verified reflective index more important than other features.