Atrial Fibrillation Detection with Low Signal-to-Noise Ratio Data Using Artificial Features and Abstract Features

Table 7

Comparison of classification results.

Author	Year	Database	Feature extraction	Task	Method	Acc	F1p	F1
Datta et al. [6]	2017	CinC 2017 AF DB	HRV, frequency domain, and statistical features	4-Class	Multilayer cascaded binary classifiers	—	—	0.830
Cao et al. [23]	2020	CinC 2017 AF DB	Abstract features	3-Class	2-Layer LSTM	0.844	—	0.827
Zabihi et al. [32]	2017	CinC 2017 AF DB	Time domain, frequency domain, time-frequency domain, and nonlinear features	4-Class	Random forest	—	0.504	0.830
Kropf et al. [33]	2017	CinC 2017 AF DB	Time-domain and frequency-domain features	4-Class	Random forest	—	0.648	0.830
Wang et al. [35]	2020	CinC 2017 AF DB	Abstract features	3-Class	DMSFNet	—	—	0.841
Gao et al. [36]	2021	CinC 2017 AF DB	Abstract features	3-Class	RTA-CNN	0.851	—	—
Mahajan et al. [37]	2017	CinC 2017 AF DB	Time domain, frequency domain, linear, and nonlinear features	4-Class	Random forest	—	—	0.780
Xiong et al. [38]	2017	CinC 2017 AF DB	Abstract features	4-Class	CNN	—	—	0.820
Zihlmann et al. [39]	2017	CinC 2017 AF DB	Abstract features	4-Class	CNN + LSTM	0.823	0.645	0.820
Gliner and Yanav [28]	2018	CinC 2017 AF DB	Time-frequency domain, statistical features, and morphological features	4-Class	SVM	—	—	0.800
Athif et al. [40]	2018	CinC 2017 AF DB	Statistical features and morphological features	4-Class	SVM	—	—	0.780
Chen et al. [41]	2018	CinC 2017 AF DB	Morphological features and heart rate variability features	4-Class	XGBoost	—	—	0.810
This work	2022	CinC 2017 AF DB	Time domain, interval, frequency domain, and nonlinear features and abstract features	4-Class	Fusion features + random forest	0.857	0.735	0.837