Interpatient ECG Heartbeat Classification with an Adversarial Convolutional Neural Network

Table 4

Performance comparison between the previous works with ours on DS2.

Work	Year	Method
[3]	2017	Features: temporal vector cardiogram + complex network	57.1%	70.7%	63.9%
		Classifier: SVM
[17]	2018	Features: features by sparse decomposition	60.8%	83.8%	72.3%
		Classifier: least-square twin SVM
[23]	2019	Multiscale CNN + RR features + beat-to-beat correlation	50.7%	92.6%	71.7%
[4]	2019	Features: wavelets + local binary patterns + higher-order statistics	60.7%	94.3%	77.5%
		+amplitude values
		Classifier: SVMs
[24]	2020	Multiperspective CNN + symbol representations + RR features	76.5%	89.7%	83.1%
[35]	2021	Features: signal morphology + higher-order statistics	52.2%	90.8%	71.5%
		+RR features
		Classifier: linear discriminant
Proposed		Adversarial CNN + RR features	84.4%	93.4%	88.9%

score is the average value of and for pathological classes S and V, defined as equation (4).