Figure 4
From: Post-stroke respiratory complications using machine learning with voice features from mobile devices
Feature importance analysis. Feature importance analysis from the XGBoost with plots demonstrating that APQ11Shimmer and RAP values are the major features even after including the clinical variables in (a) classifying those with tube feeding and (b) at risk of respiratory complications. XGBoost extreme gradient boosting, RAP relative average perturbation, APQ amplitude perturbation quotient, HNR harmonic to noise ratio, F0 fundamental frequency, MBI modified barthel index, NIHSS national institutes of health stroke scale.
