Table 6.

Comparison of different cross-validation methods and different performance metrics (column) for different models on four sensor combinations (row).

Sensors	Model	k-fold		leave-recordings-out		leave-one-subject-out
		Acc	F1	Acc	F1	Acc	F1
{LW, RW, ST, LF}	CNN-LSTM	0.45 (±0.03)	0.41 (±0.03)	0.44 (±0.04)	0.4 (±0.03)	0.44 (±0.03)	0.4 (±0.03)
{LW, RW, ST, LF}	ResNet	0.36 (±0.07)	0.35 (±0.08)	0.36 (±0.04)	0.35 (±0.05)	0.36 (±0.05)	0.35 (±0.06)
{LW, RW, ST, LF}	DeepConvLSTM	0.33 (±0.05)	0.27 (±0.06)	0.33 (±0.04)	0.25 (±0.05)	0.35 (±0.05)	0.27 (±0.05)
{LW, RW, ST, RF}	CNN-LSTM	0.44 (±0.03)	0.39 (±0.03)	0.44 (±0.03)	0.4 (±0.04)	0.43 (±0.04)	0.4 (±0.04)
{LW, RW, ST, RF}	ResNet	0.33 (±0.06)	0.32 (±0.07)	0.33 (±0.04)	0.32 (±0.05)	0.35 (±0.05)	0.33 (±0.05)
{LW, RW, ST, RF}	DeepConvLSTM	0.33 (±0.02)	0.26 (±0.02)	0.32 (±0.08)	0.26 (±0.08)	0.35 (±0.02)	0.28 (±0.02)
{LW, RW, LF, RF}	CNN-LSTM	0.43 (±0.04)	0.4 (±0.03)	0.44 (±0.04)	0.4 (±0.03)	0.44 (±0.02)	0.4 (±0.02)
{LW, RW, LF, RF}	ResNet	0.35 (±0.03)	0.34 (±0.05)	0.36 (±0.05)	0.35 (±0.05)	0.36 (±0.05)	0.34 (±0.05)
{LW, RW, LF, RF}	DeepConvLSTM	0.34 (±0.02)	0.27 (±0.03)	0.34 (±0.07)	0.27 (±0.07)	0.35 (±0.05)	0.29 (±0.06)
{LW, ST, LF, RF}	CNN-LSTM	0.41 (±0.04)	0.38 (±0.04)	0.42 (±0.04)	0.38 (±0.04)	0.41 (±0.06)	0.38 (±0.06)
{LW, ST, LF, RF}	ResNet	0.33 (±0.06)	0.32 (±0.06)	0.33 (±0.04)	0.32 (±0.04)	0.3 (±0.05)	0.29 (±0.04)
{LW, ST, LF, RF}	DeepConvLSTM	0.32 (±0.04)	0.26 (±0.03)	0.32 (±0.03)	0.26 (±0.03)	0.32 (±0.05)	0.26 (±0.06)
{RW, ST, LF, RF}	CNN-LSTM	0.39 (±0.07)	0.34 (±0.07)	0.38 (±0.02)	0.33 (±0.04)	0.38 (±0.04)	0.33 (±0.03)
{RW, ST, LF, RF}	ResNet	0.33 (±0.06)	0.31 (±0.06)	0.33 (±0.06)	0.32 (±0.07)	0.29 (±0.01)	0.28 (±0.02)
{RW, ST, LF, RF}	DeepConvLSTM	0.32 (±0.05)	0.24 (±0.08)	0.33 (±0.02)	0.25 (±0.02)	0.3 (±0.04)	0.23 (±0.04)

The best performing model for each sensor and validation setup is highlighted in bold.