Table 3.

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

Sensors	Model	k-fold		leave-recordings-out		leave-one-subject-out
		Acc	F1	Acc	F1	Acc	F1
{LW}	CNN-LSTM	0.36 (±0.03)	0.31 (±0.04)	0.36 (±0.04)	0.3 (±0.03)	0.36 (±0.04)	0.31 (±0.02)
{LW}	ResNet	0.34 (±0.02)	0.32 (±0.03)	0.33 (±0.06)	0.3 (±0.07)	0.35 (±0.04)	0.33 (±0.05)
{LW}	DeepConvLSTM	0.3 (±0.02)	0.18 (±0.04)	0.3 (±0.02)	0.19 (±0.02)	0.29 (±0.03)	0.18 (±0.03)
{RW}	CNN-LSTM	0.35 (±0.04)	0.28 (±0.05)	0.35 (±0.04)	0.28 (±0.05)	0.36 (±0.03)	0.28 (±0.03)
{RW}	ResNet	0.34 (±0.07)	0.32 (±0.07)	0.38 (±0.07)	0.36 (±0.07)	0.33 (±0.05)	0.32 (±0.04)
{RW}	DeepConvLSTM	0.27 (±0.02)	0.16 (±0.01)	0.29 (±0.04)	0.16 (±0.04)	0.29 (±0.04)	0.17 (±0.03)
{ST}	CNN-LSTM	0.32 (±0.04)	0.25 (±0.03)	0.3 (±0.03)	0.23 (±0.04)	0.32 (±0.03)	0.24 (±0.04)
{ST}	ResNet	0.26 (±0.05)	0.24 (±0.05)	0.26 (±0.06)	0.24 (±0.05)	0.29 (±0.06)	0.27 (±0.06)
{ST}	DeepConvLSTM	0.28 (±0.06)	0.19 (±0.05)	0.27 (±0.05)	0.16 (±0.05)	0.28 (±0.06)	0.18 (±0.07)
{LF}	CNN-LSTM	0.29 (±0.04)	0.23 (±0.06)	0.29 (±0.03)	0.23 (±0.04)	0.29 (±0.02)	0.23 (±0.03)
{LF}	ResNet	0.21 (±0.03)	0.19 (±0.04)	0.22 (±0.03)	0.2 (±0.03)	0.2 (±0.02)	0.18 (±0.02)
{LF}	DeepConvLSTM	0.24 (±0.03)	0.15 (±0.01)	0.21 (±0.03)	0.13 (±0.02)	0.26 (±0.05)	0.17 (±0.05)
{RF}	CNN-LSTM	0.32 (±0.02)	0.25 (±0.03)	0.3 (±0.04)	0.24 (±0.05)	0.29 (±0.04)	0.23 (±0.04)
{RF}	ResNet	0.23 (±0.06)	0.19 (±0.04)	0.25 (±0.05)	0.22 (±0.04)	0.25 (±0.05)	0.22 (±0.05)
{RF}	DeepConvLSTM	0.25 (±0.06)	0.16 (±0.06)	0.27 (±0.04)	0.18 (±0.04)	0.26 (±0.03)	0.17 (±0.03)

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