A Novel Clustering-Based Algorithm for Continuous and Noninvasive Cuff-Less Blood Pressure Estimation

Table 5

Comparison with other works.

		Machine-learning comparison (DBP)			Machine-learning comparison (SBP)
Studies	Method	MAE	RMSE	r	MAE	RMSE	r
Our proposed method	Clustering and gradient boosting regression	2.23	5.01	0.94	2.56	5.63	0.88
Our proposed method	Gradient boosting regression without clustering	6.27	10.22	0.71	6.36	10.39	0.67
[16]	SVM	6.34	—	—	12.38	—	—
[9]	Adaboosting	5.35	—	0.48	11.17	—	0.59
[17]	MLR	2.82	—	0.97	2.83	—	0.96
[27]	LSTM and perceptron	—	6.49			7.86
[11]	Multisensor features	4.54	—	0.90	6.13	—	0.84
[13]	PPG + CNN-regression	3.45	—	0.89	5.73	—	0.93
[18]	PTT + PIR + nonlinear regression	3.18	—	0.88	4.09	—	0.91
[28]	—	3.27	—	0.87	4.46	—	0.93
[29]	(PPG + ECG)	4.44	—	0.84	4.71	—	0.89
[30]	SVM	3.36	—	0.82	11.86	—	0.69
[31]	MLP	4.96	—	0.70	5.46	—	0.87
[32]	ECG: wrist and foot PPG: Finger	4.4	—	—	6.0	—	—
[33]	ANN with 15 hidden neurons	Not mentioned	—	—	3.03	—	—
[34]	PTT and PIR, regression-MARS	4.86	—	0.93	7.83	—	0.95
[25]	AutoML (TPOT)	4.19	—	—	6.52	—	—
[12]	ANN	2.21 ± 2.09	—	—	3.80 ± 3.46	—	—
[26]	DNN	6.88	—	—	9.43	—	—
[35]	Res-LSTM	4.61	—	0.74	7.10	—	0.96
[36]	LSTM-based autoencoder	4.05	—	—	2.41	—	—