Skip to main content
. 2023 Feb 20;14:1072273. doi: 10.3389/fphys.2023.1072273

TABLE 4.

Classification performance of the proposed machine learning method and deep learning method and feature-based methods on the same recordings from the MIMIC-III database. Note, NT, PHT, and HT refer to normotension, prehypertension, and hypertension, respectively. PAT stands for pulse arrival time, CWT stands for continuous wavelet transform, and KNN stands for k-nearest neighbors.

Trial Feature Classifier F1 (%)
This study (Only PPG) NT vs. PHT Tsfresh feature extraction LightGBM 90.18
NT vs. HT Tsfresh feature extraction LightGBM 97.51
(NT + PHT) vs. HT Tsfresh feature extraction LightGBM 92.77
Based on PAT and PPG features (ECG & PPG) Liang et al. (2018b) NT vs. PHT PAT and 10 PPG features KNN 84.34
NT vs. HT PAT and 10 PPG features KNN 94.84
(NT + PHT) vs. HT PAT and 10 PPG features KNN 88.49
Based on CWT and GoogLeNet (Only PPG) Liang et al. (2018a) NT vs. PHT CWT scalogram GoogLeNet 80.52
NT vs. HT CWT scalogram GoogLeNet 92.55
(NT + PHT) vs. HT CWT scalogram GoogLeNet 82.95