Fusion of Deep Features from 2D-DOST of fNIRS Signals for Subject-Independent Classification of Motor Execution Tasks

Table 13

Performance comparison between the proposed method and others.

Authors	Method	Accuracy
Siddique and Mahmud [26]	Average changes of HbO and HbR, Bayesian neural network	86.44% (2 classes)
Nazeer et al. [27]	Difference between HbO and HbR changes, LDA	98.7% (2 classes)
		85.4% (3 classes)
Bak et al. [7]	Average changes of HbO and HbR concentrations, SVM	84.4% (2 classes)
		70.4% (3 classes)
Wang et al. [28]	The transformer self-attention mechanism	75.49% (3 classes)
Wang et al. [29]	fNIRSnet	64.43% (3 classes)
Shin [30]	LIME, SVM	86.0% (2 classes)
Wang et al. [31]	GADF	78.22% (3 classes)
Proposed method	2D-DOST, feature fusion, CNN (five-fold cross-validation)	99.07% (2 classes)
		93.60% (3 classes)
Proposed method	2D-DOST, feature fusion, CNN (cross-subject cross-validation)	98.73% (2 classes)
		93.04% (3 classes)

The bold values represent the maximum accuracy.